Airport terminal cooling load prediction method and related device

Abstract

The invention belongs to the technical field of building cooling load prediction, and discloses an airport terminal cooling load prediction method and a related device, and the method comprises the steps: collecting main influence index real-time data affecting the terminal cooling load prediction; inputting the collected main influence index real-time data influencing the airport terminal cooling load prediction into the ISSA-GRNN-GRU hybrid prediction model, and outputting to obtain an airport terminal cooling load prediction result; wherein the construction process of the ISSA-GRNN-GRU hybrid prediction model comprises the following steps: acquiring influence constituent element data of terminal building cold load prediction, and training and combining a pre-constructed generalized regression neural network and a pre-constructed gating circulation unit neural network by adopting an improved sparrow search algorithm to obtain the ISSA-GRNN-GRU hybrid prediction model; the overall prediction error can be reduced, and the reliability of the prediction result is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of building cooling load prediction, and in particular relates to a cooling load prediction method for an airport terminal building and a related device. Background Art

[0002] Building cooling load refers to the heat that the air conditioning system needs to take away from the room in order to maintain the hot and humid environment inside the building and keep the indoor temperature within a certain range; the implementation of accurate building cooling load prediction is crucial to guiding the design of water storage cooling system and planning the cooling ratio, and can also improve the operating performance of the entire system, that is, by controlling the number of operating equipment, start and stop time, etc., to reduce the grid load during peak power consumption as much as possible and improve the efficiency of social energy use. However, affected by the chaos of various outdoor meteorological parameters and the status of indoor personnel, the air conditioning load presents a nonlinear characteristic of random characteristics such as dynamic and multi-disturbance, making it very difficult to solve the problem of accurate load prediction in large public buildings, especially airport terminals.

[0003] With the advancement of artificial intelligence technology, the existing building cooling load prediction methods can avoid the above problems well and demonstrate their reliability and stability in the field of building load prediction; among them, the existing building cooling load prediction methods include physical methods and data-driven methods, and the above two methods correspond to white box models and black box models, respectively.

[0004] Specifically, the physical method is highly dependent on the thermal physical properties of the building itself, and the air conditioning operation mechanism needs to be derived through physical formulas; simulation software based on physical models includes EnergyPlus, TRNSYS, ESP-r, Equest, etc., which can be modeled according to the detailed information and environmental parameters of the system; however, developing an accurate physical model is very time-consuming, and the relevant parameters are not easy to measure. The modeling process is complex and inefficient, and it is easy to have large deviations for complex buildings such as airports. In recent years, with the development of building automation control systems and Internet of Things technologies, a large amount of historical operating data of air conditioning systems has been retained; data-driven methods train data-driven models by mining historical operating data during the training process, which has the advantages of simple modeling process and high accuracy; existing data-driven methods include regression analysis prediction models, machine learning models, and deep learning prediction models.

[0005] Among them, regression analysis prediction methods include multiple linear regression, autoregression, autoregression with external input, and multiple nonlinear regression; Ravichandran mainly focuses on the impact of building materials and envelope structures on cooling load regarding building geometry, quantifies four morphological prediction factors: number of floors, exposed surface area, air-conditioned areas per building, and air-conditioned areas per floor, and evaluates cooling load on 75 major residential buildings; the results of multiple linear regression show that the four prediction factors explain 78% of the variation in cooling load; although the regression analysis prediction model has the advantages of being simpler and having low computational complexity, it cannot quickly adapt to the dynamic nonlinear changes of data, and its prediction accuracy is very limited, and is considered to be less accurate in prediction results.

[0006] In recent years, methods including artificial neural networks and support vector machines have been proven to be effective in the field of building load forecasting; Li et al. developed five machine learning models based on linear kernel support vector regression, polynomial kernel support vector regression, Gaussian radial basis function kernel support vector regression, linear regression and artificial neural network to predict the cooling load intensity of residential buildings; the results show that data-driven machine learning prediction models can provide assistance for decision-making in complex building design and renovation processes; but prediction models based on deep learning, such as neural networks, have become increasingly prominent due to their powerful nonlinear mapping and generalization capabilities, and have been favored by many researchers and scholars and applied to different prediction fields. However, the single network model parameter setting has the defect of slow convergence speed. Therefore, many scholars have introduced meta-heuristic algorithms to optimize the parameters of neural networks to improve the prediction accuracy and performance of neural network prediction models.

[0007] In the process of establishing the load forecasting model, as far as a single neural network is concerned, it has disadvantages such as difficulty in parameter setting, slow training speed, and poor generalization potential. In addition, some intelligent optimization algorithms have disadvantages such as complex calculations, insufficient stability, and easy to fall into local optimality. Therefore, it is urgent to solve the complex neural network parameter setting and nonlinear optimization problems. Summary of the invention

[0008] In view of the technical problems existing in the prior art, the present invention provides a method and related devices for predicting the cooling load of an airport terminal to solve the technical problems of complex neural network parameter setting and nonlinear optimization.

[0009] In order to achieve the above object, the technical solution adopted by the present invention is: The present invention provides a method for predicting cooling load of an airport terminal, comprising: Collect real-time data of the main influencing indicators that affect the terminal building's cooling load forecast; The collected real-time data of the main influencing indicators affecting the terminal building cooling load prediction are input into the ISSA-GRNN-GRU hybrid prediction model, and the cooling load prediction results of the airport terminal building are output; The construction process of the ISSA-GRNN-GRU hybrid prediction model is as follows: Obtaining data on influencing constituent elements of the terminal building's cooling load prediction, performing feature screening on the data on influencing constituent elements of the terminal building's cooling load prediction, and obtaining an optimal training data set; randomly splitting the optimal training data set to obtain a training subset and a test subset; Using the training subset, the pre-constructed generalized regression neural network and the pre-constructed gated recurrent unit neural network are trained using an improved sparrow search algorithm to obtain an ISSA-GRNN model and an ISSA-GRU model; Using the test subset to predict the ISSA-GRNN model and the ISSA-GRU model, and obtaining the average relative error of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment; Comparing the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment, and determining the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model; According to the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model, the ISSA-GRNN model and the ISSA-GRU model are combined to obtain an ISSA-GRNN-GRU hybrid prediction model.

[0010] Furthermore, the influencing constituent element data of the terminal cooling load prediction includes meteorological parameters affecting the airport terminal cooling load and cooling load parameters affecting the airport terminal at the current moment; Among them, the meteorological parameters affecting the cooling load of the airport terminal include outdoor dry-bulb temperature, air dew point temperature, outdoor wet-bulb temperature, relative humidity, wind speed, wind direction, solar irradiance at the previous moment and solar irradiance at the current moment; the cooling load parameters affecting the airport terminal at the current moment include the cooling load at the previous moment of the day.

[0011] Furthermore, the data of the influencing factors of the terminal building's cooling load prediction are subjected to feature screening, and in the process of obtaining the optimal training data set, the Spearman correlation coefficient method and the iterative XGBoost method are used for feature selection.

[0012] Furthermore, the improved sparrow search algorithm introduces a chaotic sequence method into the sparrow search algorithm, introduces the speed concept in the particle swarm algorithm into the finder's position update formula, introduces a nonlinear decreasing inertia weight into the finder's speed update formula, combines the following strategy of the fierce wolf in the wolf pack algorithm to update the follower's position, and introduces an adaptive inertia weight to adjust the follower's position.

[0013] Furthermore, the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment are specifically hourly cooling load prediction results for the next day.

[0014] Furthermore, the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment are compared to determine the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model, which is specifically as follows: The average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment are compared; if the average relative error of the cooling load prediction results of the ISSA-GRNN model at a certain moment is smaller than the average relative error of the cooling load prediction results of the ISSA-GRU model at a certain moment, then the moment is determined to be the operating period of the ISSA-GRNN model; otherwise, the moment is the operating period of the ISSA-GRU model.

[0015] The present invention also provides an airport terminal cooling load prediction system, comprising a real-time data acquisition module and a prediction output module; Real-time data collection module, used to collect real-time data of the main influencing indicators affecting the terminal building cooling load prediction; The prediction output module is used to input the collected real-time data of the main influencing indicators affecting the terminal building cooling load prediction into the ISSA-GRNN-GRU hybrid prediction model, and output the cooling load prediction results of the airport terminal building; The construction process of the ISSA-GRNN-GRU hybrid prediction model is as follows: Obtaining data on influencing constituent elements of the terminal building's cooling load prediction, performing feature screening on the data on influencing constituent elements of the terminal building's cooling load prediction, and obtaining an optimal training data set; randomly splitting the optimal training data set to obtain a training subset and a test subset; Using the training subset, the pre-constructed generalized regression neural network and the pre-constructed gated recurrent unit neural network are trained using an improved sparrow search algorithm to obtain an ISSA-GRNN model and an ISSA-GRU model; Using the test subset to predict the ISSA-GRNN model and the ISSA-GRU model, and obtaining the average relative error of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment; Comparing the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment, and determining the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model; According to the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model, the ISSA-GRNN model and the ISSA-GRU model are combined to obtain an ISSA-GRNN-GRU hybrid prediction model.

[0016] The present invention also provides an airport terminal cooling load prediction device, comprising: a processor suitable for executing a computer program; A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by the processor, the airport terminal cooling load prediction method is executed.

[0017] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the airport terminal cooling load prediction method is implemented.

[0018] The present invention also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the method for predicting the cooling load of an airport terminal is implemented.

[0019] Compared with the prior art, the present invention has the following beneficial effects: The airport terminal cold load prediction method provided by the present invention obtains the optimal training data set through feature screening, determines the running time of the ISSA-GRNN model and the ISSA-GRU model according to the performance of the training subset and the test subset, and selects the model with better performance for prediction by comparing the average relative errors of the cold load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment, so as to obtain the ISSA-GRNN-GRU hybrid prediction model, which can reduce the overall prediction error and improve the reliability of the prediction result; wherein, the improved sparrow search algorithm (ISSA) is used to train the GRNN and the GRU, which can efficiently search for the optimal solution, and then more effectively adjust the model parameters of the GRNN and the GRU, so that the model is more in line with the actual data and the prediction performance is improved; the present invention can greatly improve the prediction accuracy and training speed by combining the generalized regression neural network, the gated recurrent unit neural network and the improved sparrow search algorithm, reduce the complexity of the model parameters, and realize the rapid and accurate acquisition of the model parameters, which not only improves the accuracy and adaptability of the cold load prediction, but also provides strong data support for the decision-making of the airport management.

[0020] The airport terminal cooling load prediction system, airport terminal cooling load prediction device, computer-readable storage medium and computer program product provided by the present invention have all the advantages of the above-mentioned airport terminal cooling load prediction method. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart for constructing the ISSA-GRNN-GRU hybrid prediction model in Example 1; Figure 2 This is a flow chart of feature screening of the data of the influencing components of the terminal building cooling load prediction in Example 1; Figure 3 is a structural block diagram of the generalized regression neural network in Example 1; Figure 4 is a structural block diagram of the gated recurrent unit neural network in Example 1; Figure 5 It is a schematic diagram of the comparison results of the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment in Example 1; Figure 6 It is a comparison chart of the prediction results of different models in Example 1; Figure 7 A structural block diagram of the airport terminal cooling load prediction system provided in Example 2; Figure 8 This is a structural block diagram of the airport terminal cooling load prediction device provided in Example 3. DETAILED DESCRIPTION

[0022] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention more clearly understood, the present invention is further described in detail in the following specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0023] Example 1 This embodiment 1 provides a method for predicting cooling load of an airport terminal, comprising the following steps: Step 1: Construct an ISSA-GRNN-GRU hybrid prediction model, wherein the output of the ISSA-GRNN-GRU hybrid prediction model is the hourly cooling load prediction result for the next day.

[0024] As attached Figure 1 As shown, the construction process of the ISSA-GRNN-GRU hybrid prediction model is as follows: Step 11, obtain the influencing component data of the terminal cooling load prediction, which is used to determine the input and output variables of the ISSA-GRNN-GRU hybrid prediction model. The influencing component data of the terminal cooling load prediction include meteorological parameters affecting the cooling load of the airport terminal and the cooling load parameters affecting the airport terminal at the current moment; specifically, the meteorological parameters affecting the cooling load of the airport terminal include outdoor dry bulb temperature, air dew point temperature, outdoor wet bulb temperature, relative humidity, wind speed, wind direction, solar irradiance at the previous moment, and solar irradiance at the current moment; the cooling load parameters affecting the airport terminal at the current moment include the cooling load at the previous moment of the day.

[0025] Step 12: Using the Spearman correlation coefficient method and the iterative XGBoost method, feature screening is performed on the data of the influencing components of the terminal building cooling load prediction to obtain the optimal training data set.

[0026] It should be noted that there are many types of data affecting the terminal building cooling load prediction, but not all of the data have a strong correlation with the cooling load. When there are too many input variables, the accuracy of the prediction model will be reduced. Figure 2 As shown, in this embodiment 1, the Filter method and the Embedded method are integrated to perform feature screening on the data of the influencing elements of the terminal cooling load prediction; in the first stage, the Spearman correlation coefficient method in the Filter is used to preliminarily eliminate the features with low correlation, and obtain the previous In the second stage, the improved XGBoost method in Embedded is used to screen the candidate subsets and select the top The optimal features can be used to avoid the curse of dimensionality, reduce interference noise, and improve the prediction effect of the model.

[0027] (1) In the first stage, the principle of the Spearman rank correlation coefficient method in the filtering feature selection method is as follows: The Spearman rank correlation coefficient method in the filtering feature selection method is used to measure the dependence between each input variable and the dependent variable, and the features with strong correlation with the terminal cooling load are screened out from the data set to filter out irrelevant features. Indicates; for example: for two groups containing A random variable with samples and random variables , the Spearman rank correlation coefficient between them is expressed as shown in the following formula (1).

[0028] (1) in, is the Spearman rank correlation coefficient; is a random variable The samples; is a random variable The samples; is a random variable The sample mean in ; is a random variable The sample mean in .

[0029] In general, the connection between variables is insignificant, and the Spearman rank correlation coefficient can be simply calculated by the rank difference between the two variables. , as shown in formulas (2) and (3); Spearsman correlation coefficient The value range is [-1,1]; if the random variable With the random variable The Spearsman correlation coefficient increases with the increase of If it is positive, it is considered that the two groups of variables are positively correlated; otherwise, they are negatively correlated, and the Spearsman correlation coefficient The larger the absolute value of , the stronger the correlation between the variables.

[0030] (2) (3) in, for and The rank difference between them in ascending order.

[0031] It should be noted that arranging the feature selection results of the Spearman method in the first stage in descending order can effectively eliminate weakly correlated and irrelevant feature variables; however, the Spearman method realizes feature selection from a single variable perspective and fails to consider the influence between the features. If there are too many redundant features, the computational complexity of the model will increase and the interpretability of the model will be reduced; therefore, in the second stage, the iterative XGBoost method is used to select candidate input variables, thereby reducing the redundancy between the input variables.

[0032] (2) In the second stage, the principle of the extreme gradient boosting (XGBoost) algorithm is as follows: The core of the extreme gradient boosting algorithm mainly adopts the integration idea: multiple weak learners are integrated into a strong classifier through a specific method; it can not only relatively directly obtain the importance score of each feature and realize the feature selection function, but also for samples with missing feature values, XGBoost can automatically learn the splitting direction of the feature; it can also support custom loss functions, and use the second-order Taylor formula to expand and optimize the loss function, reduce deviation, and improve calculation accuracy; and add regularization terms to the objective function to reduce the model variance, simplify the complexity of the algorithm to prevent overfitting, and then the accuracy of the model is jointly determined by the deviation and variance; among them, the added regularization terms include the number of leaf nodes in the tree and the L2 norm of the leaf node weights.

[0033] Specifically, the objective function consists of the model's loss function and a regularization term that controls the model's complexity, and is defined as shown in the following formula (4).

[0034] (4) in, is the loss function calculated by the true value and the predicted value, representing the deviation of the model; is the number of trees; is the function space of all trees; is the regularization term.

[0035] According to the tree integration model and gradient method, the approximate objective function after the second-order Taylor formula is expanded is shown in the following formula (5).

[0036] (5) in, is the current iteration number; For the newly added tree; is the first-order derivative of the loss function, is the second-order derivative of the loss function; is the total number of leaf nodes; is the penalty coefficient; is the weight of the leaf node.

[0037] In order to make the model more simple, remove the constant term and continue to process the objective function, let , , and the final objective function is obtained; wherein the final objective function is shown in the following formula (6).

[0038] (6) in, I is the sample set of each leaf node, ; The structure of the tree.

[0039] After determining the structure of the tree, the optimal weight of the leaf node and the optimal objective function value can be obtained, as shown in the following formula (7).

[0040] (7) If the objective function value is smaller, it means that the tree structure is better. At this time, the global importance of the feature and the individual importance score are calculated as follows (8).

[0041] (8) in, Characterized by K The average importance score obtained in the trees; is the importance score of the feature in a single tree; is the square loss reduction after the node splits.

[0042] It should be noted that the iterative XGBoost feature selection method is to combine the XGBoost algorithm with the iterative strategy to perform a second feature selection on the candidate subset after division to remove redundant features; specifically, the features obtained in the first stage are The features are divided into The XGBoost method is used to calculate the feature importance of each feature subset in turn; the features with higher scores in each feature subset are selected and put into the optimal feature subset, and the optimal feature subset is added to the next feature subset to recalculate the importance score; the above process is repeated, and the traversal The iterative XGBoost feature selection method can not only retain the correlation between variables, but also weaken the redundancy between features due to the small dimension processed each time, and then effectively eliminate redundant features through iterative strategies.

[0043] Step 13: randomly split the optimal training data set to obtain a training subset and a test subset.

[0044] Step 14: Using the training subset, an improved sparrow search algorithm is used to train the pre-constructed generalized regression neural network and the pre-constructed gated recurrent unit neural network to obtain an ISSA-GRNN model and an ISSA-GRU model.

[0045] Generalized Regression Neural Network (GRNN) is a second-type radial basis function neural network (RBF) proposed by Specht. It is an improvement based on the RBF network and has stronger nonlinear mapping ability and learning speed than RBF. GRNN is trained without iterative solutions, so it has a faster training speed and good generalization ability, robustness and fault tolerance. It can better handle unstable data such as building cooling load and is widely used in various disciplines and engineering fields such as time series prediction, classification and control decision systems.

[0046] As attached Figure 3 As shown, the generalized regression neural network includes an input layer, a pattern layer, a summation layer and an output layer.

[0047] In the input layer, the number of neurons in the input layer is equal to the dimension of the input vector in the training subset, and each neuron directly transmits the input variable to the pattern layer.

[0048] In the pattern layer, the number of neurons in the pattern layer is proportional to the number of training samples. n are equal; wherein the neuron transfer function in the pattern layer is as shown in the following formula (9).

[0049] (9) in, Input samples to the network; For the The training samples corresponding to neurons; is the smoothing factor; is the output of the neuron.

[0050] In the summation layer, there are two different types of neurons; the first type of neuron calculates the algebraic sum of the outputs of all neurons in the pattern layer; the connection weight between the pattern layer and the first type of neuron is 1, and the transfer function is shown in the following formula (10); the second type of neuron performs weighted summation of all neurons in the pattern layer, that is, the first type of neuron in the pattern layer The neurons in the summation layer The sum of the molecules is the connection weight between neurons. Output samples The elements, the transfer function is shown in formula (11).

[0051] (10) (11) in, The algebraic sum result output by the sum layer; It is the weighted summation result output by the summation layer.

[0052] In the output layer, the number of neurons in the output layer is equal to the dimension of the output vector in the training sample ,but Divide the output of the summing layer for each neuron as shown in formula (12).

[0053] (12) in, It is the division result output by the output layer.

[0054] It should be noted that the performance of GRNN depends on factors such as feature selection and model parameters; among them, the smoothing factor in the pattern layer The performance of the network is greatly affected; in this embodiment 1, in order to improve the prediction accuracy and generalization ability of GRNN, an improved sparrow search algorithm is used to smooth the factor in the pattern layer. Conduct optimization.

[0055] The Gate Recurrent Unit (GRU) is a recurrent neural network (RNN) for processing time series. GRU is derived from the Long-Short Term Memory (LSTM) neural network. It integrates the forget gate and input gate of LSTM into a single update gate. Compared with traditional RNN, it can perform better in capturing long-term dependencies in sequence data, solving the gradient vanishing and gradient exploding problems of traditional RNN while maintaining a low computational cost. Compared with the LSTM model, it has fewer parameters, which can greatly improve training efficiency and prevent overfitting. Therefore, it has a wide range of applications in natural language processing, speech recognition, time series prediction and other fields.

[0056] As attached Figure 4 As shown, the gated recurrent unit neural network includes two gate functions: a reset gate and an update gate; the forward propagation formula of the gated recurrent unit neural network is shown in the following formulas (13)-(17).

[0057] (13) (14) (15) (16) (17) in, To reset the gate; To update the door; It is the hidden state transmitted from the previous moment; is the input at the current moment; is the output at the current moment; is the Sigmoid activation function; is the hyperbolic tangent activation function; is the weight matrix of the reset gate; is the weight matrix of the update gate; is the weight matrix of the hidden layer; is the weight matrix of the output gate; is the bias vector for resetting the gate; is the bias vector of the update gate; is the bias vector of the hidden layer; is the bias vector of the output gate; * is element-wise multiplication.

[0058] In this embodiment 1, an improved sparrow search algorithm is used to train a pre-constructed generalized regression neural network and a pre-constructed gated recurrent unit neural network to optimize the parameters of the generalized regression neural network and the gated recurrent unit neural network; wherein the improved sparrow search algorithm introduces a chaotic sequence method into the sparrow search algorithm, introduces the speed concept in the particle swarm algorithm into the finder's position update formula, introduces a nonlinear decreasing inertia weight into the finder's speed update formula, updates the follower's position in combination with the following strategy of the fierce wolf in the wolf pack algorithm, and introduces an adaptive inertia weight to adjust the follower's position.

[0059] (a) Since the sparrow search algorithm generates the initial population in a random manner, it cannot ensure that the initial positions of the individuals are evenly distributed in the search space, which in turn affects the algorithm's search speed and performance. When solving the optimal parameters of the neural network, this embodiment 1 introduces a chaotic sequence method in the initialization process of the Improved Sparrow Search Algorithm (ISSA) to ensure the uniform distribution of the initial solution.

[0060] The basic idea of ​​the chaotic sequence method is to utilize the ergodicity and randomness of chaotic variables, and under the deterministic condition of satisfying the optimization parameter range constraints, map the optimization variables to the value interval of the chaotic variables through the chaotic mapping rule, and transform the obtained chaotic sequence into the search space of the objective function through linear transformation, so that the feasible solutions are evenly distributed in the space; among them, the chaotic sequence method uses cubic mapping on the sparrow individuals, as shown in the following formula (18).

[0061] (18).

[0062] (b) In the initial stage of SSA search, the finder quickly concentrates on the position of the global optimal value, resulting in a decrease in population diversity, causing the algorithm to converge to the local optimal solution earlier, thereby causing the problem of low search accuracy; therefore, this embodiment 1 introduces the concept of speed in the particle swarm algorithm into the finder's position update formula to more effectively realize the position information communication ability of individual sparrows, thereby avoiding the algorithm from converging to the optimal solution too early; wherein, the improved finder position update formula is shown in the following formulas (19)-(20).

[0063] (19) (20) in, All are learning factors; The range is A random number; is the inertia weight; For the The sparrow individual is in Speed ​​in the dimension; For the The sparrow individual is in Dimensional position.

[0064] (c) Inertia weight It represents the ability of the individual sparrow to inherit its previous speed. A nonlinear decreasing inertia weight is added to the speed update formula of the discoverer to improve the algorithm optimization accuracy. The nonlinear decreasing inertia weight update formula is shown in the following formula (21).

[0065] (twenty one) in, is the inertia weight; is the maximum value of inertia weight; is the minimum value of the inertia weight.

[0066] (d) In the wolf pack algorithm, the fierce wolves approach the leader with a larger step size and solve the optimization problem with the "survival of the strong" update mechanism, which has better global search capabilities; while the SSA algorithm is prone to the problem of rapid decline in population diversity in the late iteration when solving the optimization problem of neural network parameters, and is trapped in the problem of local optimal solution that is difficult to jump out, thereby limiting the search efficiency of the algorithm; in this embodiment 1, the position of the follower is updated in combination with the following strategy of the fierce wolves in the wolf pack algorithm, that is, the follower quickly approaches the discoverer with a variable step size and searches for food, thereby improving the ability of the population to jump out of the local optimal solution; wherein, the improved follower position update formula is shown in the following formulas (22)-(23).

[0067] (twenty two) (twenty three) in, is the change step size.

[0068] From the above improved follower position update formula, we can see that The position of each sparrow will be determined according to the Only the position of the individual sparrow is updated, and it is highly dependent on the previous individual; if the position of the follower is a local optimal solution, it will easily fall into the local optimal solution and stagnate; in this embodiment 1, in order to better balance the search ability and development ability of the SSA algorithm in the later stage, an adaptive inertia weight is introduced to adjust the follower position, which determines the influence of the previous individual on the current individual. When the value is large, it helps improve the search ability; When the value is small, it helps to develop specific capabilities. The process of adaptive inertia weight adjusting the follower position is shown in the following formula (24).

[0069] (twenty four) in, is the adaptive inertia weight.

[0070] In this embodiment 1, an improved sparrow search algorithm is used to optimize the parameters of the GRNN neural network and the GRU neural network to prevent them from falling into local optimality and being unable to jump out, thereby ensuring the prediction accuracy and training speed of the model, and using the mean absolute error (MAE) of the GRNN and GRU neural network training sets as the fitness evaluation function of ISSA; wherein the fitness evaluation function of ISSA is shown in the following formula (25).

[0071] (25) in, is the actual value of the training subset, is the cooling load prediction value of the training subset, n is the sample size of the training subset.

[0072] Step 15, using the test subset to predict the ISSA-GRNN model and the ISSA-GRU model, and obtaining the average relative error of the cold load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment. Specifically, the test subset is input into the ISSA-GRNN model and the ISSA-GRU model, and the cold load prediction values ​​of the test subset are output respectively; according to the fitness evaluation function of the ISSA, the average relative error of the cold load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment is solved.

[0073] Step 16: Compare the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment to determine the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model; specifically, the process of determining the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model is as follows: The average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment are compared; if the average relative error of the cooling load prediction results of the ISSA-GRNN model at a certain moment is smaller than the average relative error of the cooling load prediction results of the ISSA-GRU model at a certain moment, then the moment is determined to be the operating period of the ISSA-GRNN model; otherwise, the moment is the operating period of the ISSA-GRU model.

[0074] It should be noted that in this embodiment 1, the ISSA-GRNN-GRU hybrid prediction model is verified by obtaining data from the cooling area of ​​a certain airport terminal, and the relevant parameters are adjusted according to the prediction performance results to ensure that the constructed ISSA-GRNN-GRU hybrid prediction model can accurately predict the cooling load.

[0075] Step 2: Collect real-time data of the main influencing indicators that affect the terminal building cooling load prediction.

[0076] Step 3: Input the collected real-time data of the main influencing indicators affecting the terminal cooling load prediction into the ISSA-GRNN-GRU hybrid prediction model, and output the cooling load prediction result of the airport terminal.

[0077] Example description: The performance test of the airport terminal cooling load prediction method was conducted using the data of the cooling area of ​​an airport terminal in Northwest China; the details are as follows: According to the actual working characteristics of the airport terminal building, that is, 24-hour cooling, a total of 1608 groups of hourly cooling load data from June 1 to August 6 were selected from the cooling load sample data after feature selection as the training data of the model, and a total of 168 groups of data from August 7 to 13 were used as the test data of the model; the simulation experiment was carried out on a PC with Windows 10 system and Intel Core (TM) i5-8265U processor based on the MATLAB R2021b software platform; the prediction performance of the ISSA-GRNN-GRU prediction model was verified and analyzed through multiple experiments; the relevant parameters of the ISSA-GRNN-GRU prediction model were adjusted and tested during the experiment, and finally a set of parameter settings suitable for terminal cooling load prediction were obtained, that is, the optimal model parameters of the ISSA-GRNN-GRU prediction model were obtained.

[0078] 1464 sets of data from June 1 to July 31 in the cooling load sample data set were selected as training subsets to build the ISSA-GRNN and ISSA-GRU prediction models respectively, and the remaining data in the training set were used as test subsets for verification. The average relative error of the cooling load output values ​​of the two prediction models at different times was calculated. The results are shown in Figure 2. Figure 5 As shown; from the attached Figure 5 It can be seen that at 0, 2, 4, 7, 8, 11, 12, 17, 18, and 21 moments, the average relative error predicted by the ISSA-GRNN model is smaller; therefore, the above 0, 2, 4, 7, 8, 11, 12, 17, 18, and 21 moments are used as the operating periods of the ISSA-GRNN model to use the ISSA-GRNN model for cooling load prediction; the remaining moments are predicted by the ISSA-GRU model, thereby completing the establishment of the ISSA-GRNN-GRU hybrid prediction model.

[0079] The test set was used to compare and analyze the prediction results of the five models: GRNN, GRU, ISSA-GRNN, ISSA-GRU, and ISSA-GRNN-GRU. The results of the accuracy evaluation indicators of the obtained prediction models are shown in Table 1.

[0080] Table 1 Comparison of performance evaluation results of five prediction models

[0081] It can be seen from Table 1 above that compared with the other four models, the RMSE and MAPE of the ISSA-GRNN-GRU hybrid prediction model are the lowest, which are reduced by 3.9963kW-30.5921kW and 1.1253%-3.5693% respectively; this shows that the ISSA-GRNN-GRU hybrid prediction model has higher prediction accuracy; at the same time, compared with GRNN and GRU, the error of the model after optimization by the ISSA algorithm is smaller. It can be seen that the ISSA algorithm can be effectively applied to the optimization problem of neural networks.

[0082] Furthermore, the prediction results of the five prediction models for the hourly cooling load of the terminal are shown in the form of a dot-line graph. In order to compare the prediction effects of each model more clearly, as shown in the attached figure, Figure 6 The prediction results for the first three days of the test set are shown in the figure. Figure 6 It can be seen that by establishing different models to predict at different times, the deviation between the cooling load prediction value of the ISSA-GRNN-GRU hybrid prediction model and the actual value is smaller, and even almost completely coincides with the actual value.

[0083] The airport terminal cooling load prediction method described in this embodiment 1 uses a training subset to model ISSA-GRNN and ISSA-GRU, uses the ISSA-GRNN and ISSA-GRU models to predict the test subset, and calculates the average relative error at each moment; by comparing the prediction results of the two models, the prediction time of each prediction model is determined, and the modeling of the ISSA-GRNN-GRU hybrid prediction model is completed; wherein, the improved sparrow search algorithm is used to optimize the parameters of the generalized regression neural network and the gated recurrent unit neural network, and the constructed ISSA-GRNN-GRU hybrid prediction model can obtain more accurate prediction results when predicting the cooling load, so as to guide the formulation of a multi-objective optimization strategy for the water storage air conditioning system.

[0084] Example 2 As attached Figure 7 As shown, this embodiment 2 provides an airport terminal cooling load prediction system, including a model building module, a real-time data acquisition module and a prediction output module.

[0085] The model building module is used to build the ISSA-GRNN-GRU hybrid prediction model; the real-time data acquisition module is used to collect the real-time data of the main influencing indicators that affect the terminal cooling load prediction; the prediction output module is used to input the collected real-time data of the main influencing indicators that affect the terminal cooling load prediction into the ISSA-GRNN-GRU hybrid prediction model, and output the cooling load prediction result of the airport terminal.

[0086] In this embodiment 2, the construction process of the ISSA-GRNN-GRU hybrid prediction model is as follows: The influencing constituent element data of the terminal building cooling load prediction are obtained, and the influencing constituent element data of the terminal building cooling load prediction are subjected to feature screening to obtain an optimal training data set; the optimal training data set is randomly split to obtain a training subset and a test subset; the training subset is used to train a pre-constructed generalized regression neural network and a pre-constructed gated recurrent unit neural network using an improved sparrow search algorithm to obtain an ISSA-GRNN model and an ISSA-GRU model; the test subset is used to predict the ISSA-GRNN model and the ISSA-GRU model to obtain the ISSA-GRNN model. The method comprises the following steps: calculating the average relative error of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment; comparing the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment to determine the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model; combining the ISSA-GRNN model and the ISSA-GRU model according to the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model to obtain the ISSA-GRNN-GRU hybrid prediction model.

[0087] Example 3 As attached Figure 8 As shown, this embodiment 3 provides an airport terminal cooling load prediction device, including: a memory for storing a computer program; a processor for implementing the steps of the airport terminal cooling load prediction method when executing the computer program.

[0088] When the processor executes the computer program, the steps of the above-mentioned airport terminal cooling load prediction method are implemented, for example: Construct an ISSA-GRNN-GRU hybrid prediction model; collect real-time data of the main influencing indicators that affect the terminal cooling load prediction; input the collected real-time data of the main influencing indicators that affect the terminal cooling load prediction into the ISSA-GRNN-GRU hybrid prediction model, and output the airport terminal cooling load prediction result.

[0089] Alternatively, when the processor executes the computer program, the functions of each module in the above-mentioned airport terminal cooling load prediction system are realized, for example: The model building module is used to build the ISSA-GRNN-GRU hybrid prediction model; the real-time data acquisition module is used to collect the real-time data of the main influencing indicators that affect the terminal cooling load prediction; the prediction output module is used to input the collected real-time data of the main influencing indicators that affect the terminal cooling load prediction into the ISSA-GRNN-GRU hybrid prediction model, and output the cooling load prediction result of the airport terminal.

[0090] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing preset functions, and the instruction segments are used to describe the execution process of the computer program in the airport terminal cooling load prediction device.

[0091] For example, the computer program can be divided into a model building module, a real-time data collection module and a prediction output module, and the specific functions of each module are as follows: a model building module, used to build an ISSA-GRNN-GRU hybrid prediction model; a real-time data collection module, used to collect real-time data of the main influencing indicators that affect the terminal cooling load prediction; a prediction output module, used to input the collected real-time data of the main influencing indicators that affect the terminal cooling load prediction into the ISSA-GRNN-GRU hybrid prediction model, and output the airport terminal cooling load prediction result.

[0092] The airport terminal cold load prediction device can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The airport terminal cold load prediction device can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above is an example of an airport terminal cold load prediction device and does not constitute a limitation on the airport terminal cold load prediction device. It can include more components than the above, or a combination of certain components, or different components. For example, the airport terminal cold load prediction device can also include input and output devices, network access devices, buses, etc.

[0093] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The processor is the control center of the airport terminal cooling load prediction device, and uses various interfaces and lines to connect various parts of the entire airport terminal cooling load prediction device.

[0094] The memory can be used to store the computer program and / or module, and the processor implements various functions of the airport terminal cooling load prediction device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0095] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0096] Example 4 This embodiment 4 also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for predicting the cooling load of an airport terminal are implemented, for example: According to the requirements of the configuration software, determine the image resources that need to be preloaded and obtain the preloaded images; According to the requirements of the configuration software, determine the image resources that need to be preloaded and obtain the preloaded images; Use the ImageIcon class provided by Java Swing to define a HashMap<String, ImageIcon> A private member variable imageCache of type is used to obtain a predefined local image cache; monitor whether a preset event occurs or whether a preset timing time is reached; if so, store the preloaded image in the predefined local image cache; according to the received interface display request, obtain the image resources to be displayed from the predefined local image cache, and display the image resources to be displayed; according to the preset cache size limit or the preset time interval, clean up the image resources to be displayed stored in the predefined local image cache.

[0097] If the module / unit integrated in the airport terminal cooling load prediction system is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0098] Based on such understanding, the present invention implements all or part of the processes in the above-mentioned airport terminal cooling load prediction method, and can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned airport terminal cooling load prediction method can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or preset intermediate form, etc.

[0099] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0100] It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0101] Example 5 This embodiment 5 provides a computer product, which includes a computer program, and the computer program is stored in a computer-readable storage medium; the processor of the airport terminal cooling load prediction device reads the computer program from the computer-readable storage medium, and the processor executes the computer program, so that the airport terminal cooling load prediction device can execute the airport terminal cooling load prediction method described in embodiment 1, which will not be repeated here.

[0102] It should be noted that a person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods.

[0103] The airport terminal cooling load prediction method provided by the present invention uses the Spearman correlation coefficient method in the Filter to preliminarily eliminate the features with low correlation and obtain the previous An important feature is selected to filter out irrelevant features, reduce the dimension of the input variables of the prediction model, improve the convergence speed, and save computing costs; the iterative XGBoost method is used to select candidate input variables. The advantage of the iterative XGBoost feature selection method is that it can not only retain the correlation between variables, but also because the dimension processed each time is small, it can well weaken the redundancy between features, and then effectively eliminate redundant features through iterative strategies; the parameters of the GRNN neural network and the GRU neural network are optimized through the improved sparrow search algorithm (ISSA) to prevent them from falling into the local optimum and being unable to jump out, and the mean absolute error of the GRNN and GRU neural network training sets is used as the fitness evaluation function of ISSA, which greatly improves the prediction accuracy and training speed and reduces the complexity of the data.

[0104] In the present invention, the ISSA-GRNN-GRU hybrid prediction model performs very stably in different time periods and data sets, and has good generalization ability; the deviation between the prediction results and the actual values ​​is extremely small, and even almost completely overlaps in many cases. The R² value of the ISSA-GRNN-GRU hybrid prediction model is close to 1, and its fitting degree is very high, which can accurately predict the hourly cooling load of the terminal; compared with GRNN, GRU, ISSA-GRNN, ISSA-GRU and other models, the RMSE and MAPE of the ISSA-GRNN-GRU hybrid prediction model are the lowest, which are reduced by 3.9963kW-30.5921kW and 1.1253%~3.5693% respectively, indicating that the ISSA-GRNN-GRU hybrid prediction model has obvious advantages in prediction accuracy and performance.

[0105] The above embodiment is only one of the implementation methods that can realize the technical solution of the present invention. The scope of protection claimed by the present invention is not limited only to this embodiment, but also includes changes, replacements and other implementation methods that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention.

Claims

1. A method for predicting cooling load of an airport terminal, characterized in that: include: Collect real-time data of the main influencing indicators that affect the terminal building's cooling load forecast; The collected real-time data of the main influencing indicators affecting the terminal building cooling load prediction are input into the ISSA-GRNN-GRU hybrid prediction model, and the cooling load prediction results of the airport terminal building are output; The construction process of the ISSA-GRNN-GRU hybrid prediction model is as follows: Obtaining data on influencing constituent elements of the terminal building's cooling load prediction, performing feature screening on the data on influencing constituent elements of the terminal building's cooling load prediction, and obtaining an optimal training data set; randomly splitting the optimal training data set to obtain a training subset and a test subset; Using the training subset, the pre-constructed generalized regression neural network and the pre-constructed gated recurrent unit neural network are trained using an improved sparrow search algorithm to obtain an ISSA-GRNN model and an ISSA-GRU model; Using the test subset to predict the ISSA-GRNN model and the ISSA-GRU model, and obtaining the average relative error of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment; Comparing the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment, and determining the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model; According to the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model, the ISSA-GRNN model and the ISSA-GRU model are combined to obtain an ISSA-GRNN-GRU hybrid prediction model.

2. The method for predicting cooling load of an airport terminal according to claim 1, characterized in that: The influencing constituent element data of the terminal cooling load prediction include meteorological parameters affecting the cooling load of the airport terminal and cooling load parameters affecting the airport terminal at the current moment; Among them, the meteorological parameters affecting the cooling load of the airport terminal include outdoor dry-bulb temperature, air dew point temperature, outdoor wet-bulb temperature, relative humidity, wind speed, wind direction, solar irradiance at the previous moment and solar irradiance at the current moment; the cooling load parameters affecting the airport terminal at the current moment include the cooling load at the previous moment of the day.

3. The method for predicting cooling load of an airport terminal according to claim 1, characterized in that: The data of the influencing factors of the terminal building's cooling load prediction are subjected to feature screening. In the process of obtaining the optimal training data set, the Spearman correlation coefficient method and the iterative XGBoost method are used for feature selection.

4. The method for predicting cooling load of an airport terminal according to claim 1, characterized in that: The improved sparrow search algorithm introduces a chaotic sequence method into the sparrow search algorithm, introduces the speed concept in the particle swarm algorithm into the finder's position update formula, introduces a nonlinear decreasing inertia weight into the finder's speed update formula, combines the following strategy of the fierce wolf in the wolf pack algorithm to update the follower's position, and introduces an adaptive inertia weight to adjust the follower's position.

5. The method for predicting cooling load of an airport terminal according to claim 1, characterized in that: The cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment are specifically hourly cooling load prediction results for the next day.

6. The method for predicting cooling load of an airport terminal according to claim 1, characterized in that: The process of comparing the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment to determine the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model is as follows: The average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment are compared; if the average relative error of the cooling load prediction results of the ISSA-GRNN model at a certain moment is smaller than the average relative error of the cooling load prediction results of the ISSA-GRU model at a certain moment, then the moment is determined to be the operating period of the ISSA-GRNN model; otherwise, the moment is the operating period of the ISSA-GRU model.

7. An airport terminal cooling load prediction system, characterized in that: Including real-time data acquisition module and prediction output module; Real-time data collection module, used to collect real-time data of the main influencing indicators affecting the terminal building cooling load prediction; The prediction output module is used to input the collected real-time data of the main influencing indicators affecting the terminal building cooling load prediction into the ISSA-GRNN-GRU hybrid prediction model, and output the cooling load prediction results of the airport terminal building; The construction process of the ISSA-GRNN-GRU hybrid prediction model is as follows: Obtaining data on influencing constituent elements of the terminal building's cooling load prediction, performing feature screening on the data on influencing constituent elements of the terminal building's cooling load prediction, and obtaining an optimal training data set; randomly splitting the optimal training data set to obtain a training subset and a test subset; Using the training subset, the pre-constructed generalized regression neural network and the pre-constructed gated recurrent unit neural network are trained using an improved sparrow search algorithm to obtain an ISSA-GRNN model and an ISSA-GRU model; Using the test subset to predict the ISSA-GRNN model and the ISSA-GRU model, and obtaining the average relative error of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment; Comparing the average relative errors of the cooling load prediction results of the ISSA-GRNN model and the ISSA-GRU model at each moment, and determining the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model; According to the running time of the ISSA-GRNN model and the running time of the ISSA-GRU model, the ISSA-GRNN model and the ISSA-GRU model are combined to obtain an ISSA-GRNN-GRU hybrid prediction model.

8. An airport terminal cooling load prediction device, characterized in that: include: a processor suitable for executing a computer program; A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by the processor, the method for predicting the cooling load of an airport terminal according to any one of claims 1 to 6 is executed.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting the cooling load of an airport terminal as described in any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method for predicting the cooling load of an airport terminal as claimed in any one of claims 1 to 6 is implemented.

Images