Water chilling unit load prediction method based on CVA and genetic optimization ELM fusion

Abstract

The invention provides a water chilling unit load prediction method based on CVA and genetic optimization ELM fusion, and belongs to the technical field of water chilling unit load prediction. The technical problem of excessive refrigeration or frequent start and stop is solved. According to the technical scheme, the method comprises the following steps that S1, collected open source water chilling unit data are divided into past data and future data; s2, constructing a Hankel matrix; s3, performing singular value decomposition on the Hankel matrix; s4, taking the standard variable matrix Xest as input data of a subsequent model; s5, initializing network parameters of the extreme learning machine; s6, calculating a hidden layer output matrix and a weight value matrix; and S7, respectively inputting the test data into the CVA-GA-ELM load prediction model and the PCA-BP load prediction model. The method has the beneficial effects that the adverse effects of high-dimensional and dynamic characteristics of data are reduced, the stability of the model is enhanced, and the accuracy of load prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of chiller load prediction, and in particular to a chiller load prediction method based on the fusion of CVA and genetic optimization ELM. Background Art

[0002] With the continuous development of modern society, people are increasingly dependent on air conditioning in their lives. As the main energy-consuming equipment in large public buildings, central air conditioning accounts for a large proportion of energy consumption losses. The operating conditions of its main component, the chiller, are complex, and there are many factors that need to be considered when building a prediction model involving multiple variables. General prediction models are mainly divided into three types: white box, gray box, and black box models. Among them, the white box model is mainly represented by the traditional thermal theory model. The black box model is mainly based on a data-driven inverse model, which relies on a large amount of data for model identification to achieve accurate model prediction. The gray box model is a fusion model between the white box model and the black box model. Accurate load forecasting is often crucial in formulating demand scheduling, and it also makes certain contributions to energy-saving optimization. Therefore, it is necessary to study a method to improve the accuracy of chiller load forecasting.

[0003] Canonical Variate Analysis (CVA), as a multivariate statistical analysis technique, has unique advantages in processing dynamic data characteristics. Considering the time series characteristics of data, it can better capture the dynamic relationship between variable data and provide accurate data analysis results. It is mainly used to study the relationship between variables, emphasizing both the autocorrelation of a single variable and the correlation between different variables, aiming to find linear combinations to maximize the correlation between variables. Therefore, CVA can be used to reduce the dimensionality of multivariate data. Its purpose is to reduce the number of data features while maintaining the main information of the data as much as possible, helping to reduce the complexity of the model and improve computational efficiency.

[0004] Extreme Learning Machine (ELM) is a single hidden layer feedforward neural network model. The core idea is to randomly generate input weights and biases, and then solve the output weights through the least squares method, thereby avoiding the time-consuming gradient descent process in traditional neural networks and solving problems with less execution time. It has fast learning speed, simple structure, easy implementation, and strong generalization ability. However, since the weights and biases are randomly initialized, this has a certain impact on the performance of the model, and it is impossible to achieve the global optimal effect at once, resulting in low prediction accuracy.

[0005] Genetic Algorithm (GA) is an optimization and search technology based on natural selection and genetic mechanisms. It generates better solutions from the population by simulating the biological evolution process: operations such as selection, crossover and mutation, thereby effectively avoiding falling into the local optimal solution. The algorithm does not rely on gradient information and can be applied to non-continuous and nonlinear problems. Based on this, the use of genetic algorithms to optimize the input weights and hidden layer bias parameters of the extreme learning machine can greatly improve the performance of the load forecasting model and improve the prediction accuracy. Chinese patent CN202310659069.4 is a chiller group control optimization method based on load forecasting. The patent proposes to establish a chiller load forecasting model based on the cooling load energy consumption, and train the prediction model based on the PCA-BP artificial neural network. Although this method improves the model performance to a certain extent, it lacks consideration of nonlinear and dynamic characteristics, resulting in low prediction accuracy. There are still large errors and limitations in practical applications. Therefore, how to extract effective information from the data of chillers with nonlinear and dynamic characteristics to achieve load forecasting while improving the prediction accuracy is a more challenging research problem in chiller load forecasting. Summary of the invention

[0006] The purpose of the present invention is to provide a chiller load forecasting method based on the fusion of CVA and genetic optimization ELM. The method makes full use of canonical variable analysis to reduce multi-dimensional data to low-dimensional data features, thereby retaining the data integrity to the greatest extent and combining it with an extreme learning machine. In order to further improve the prediction accuracy, a genetic algorithm is introduced to optimize the bias and weight of the extreme learning machine, and a load forecasting model CVA-GA-ELM model is constructed to obtain a higher prediction accuracy within an acceptable error range.

[0007] The inventive idea of ​​the present invention is: the present invention provides a chiller load forecasting method based on the fusion of CVA and genetic optimization ELM. CVA has unique advantages in processing dynamic data characteristics. Considering the time series characteristics of data, it better captures the dynamic relationship between variable data and can provide accurate data analysis results. The data is divided into training data and test data. The training data is first pre-processed to remove abnormal values. Then the training data is divided into past and future data, and the covariance matrix and the mutual covariance matrix between them are calculated respectively. After further constructing the Hankel matrix, the singular value decomposition is performed. After obtaining the conversion matrix, the number of main canonical variables is determined according to the cumulative contribution, thereby reducing the dimension. The reduced-dimensional data is further input into the extreme learning machine for load forecasting. The input weights and biases of the extreme learning machine are randomly generated, resulting in unstable prediction results and poor model performance. In order to optimize the prediction performance, a genetic algorithm is introduced for iterative optimization, the purpose of which is to find the optimal bias parameters and weight parameters and assign them to the extreme learning machine to obtain the CVA-GA-ELM load forecasting model. Then the test data is input into the model for chiller load forecasting to obtain the load forecasting result. The error comparison with the PCA-BP prediction model shows that the prediction effect of this model is significantly better than that of the PCA-BP model, and it has high practical application value. The present invention combines the standard variable with the genetic optimization extreme learning machine for load forecasting, which is a new attempt in the field of chiller load forecasting. It not only reduces the adverse effects of high-dimensional and dynamic characteristics of data, but also enhances the stability of the model, and also improves the accuracy of load forecasting through optimization strategies.

[0008] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is specifically: a chiller load forecasting method based on the fusion of CVA and genetic optimization ELM, comprising the following steps:

[0009] Step S1: Divide the collected open source chiller data into past data and future data, set the length p of the past observation value and the length f of the future observation value and the given system order n;

[0010] Step S2: Construct the Hankel matrix and calculate the covariance matrix Σ of past and future observations pp ,Σ ff and the cross-covariance matrix Σ fp ;

[0011] Step S3: Perform singular value decomposition on the Hankel matrix, calculate the projection matrices J and L, and obtain the reduced-dimensional canonical variable matrix X est ;

[0012] Step S4: Convert the canonical variable matrix X estAs the input data of the subsequent model, it is divided into training data p1, t1 and test data p2, t2, and data normalization is performed;

[0013] Step S5: Initialize the extreme learning machine network parameters, set the number of neurons in the hidden layer of the ELM, set the optimization target, initialize the population and optimization parameters, calculate the fitness function of the GA algorithm, and after meeting the optimization target, use the optimized parameters as the weight value and bias value between the input layer and the hidden layer of the ELM;

[0014] Step S6: Calculate the hidden layer output matrix and weight value matrix, input the training data p1 and t1 into the model training, and use the test data p2 and t2 for verification to obtain the CVA-GA-ELM load forecasting model, and obtain the load forecasting accuracy within the acceptable error range;

[0015] Step S7: Send the same data into the PCA-BP prediction model, obtain the prediction result, compare the error with the present invention, and obtain the final conclusion.

[0016] Further, in step S1, the collected open source chiller data is divided into past data y p(r) and future data f(r) , where r is an ordinal number, and p and f refer to the lengths of past and future observations, respectively.

[0017]

[0018]

[0019] Furthermore, in step S2, the Hankel matrix Y is constructed p ∈R,Y f ∈R, which is defined as follows: the maximum number of columns of the Hankel matrix is ​​N=lp-f+1, where l is the observed variable.

[0020] Y p =[y p(p+1) y p(p+2) …y p(p+N) ] (3)

[0021] Y f =[y f(p+1) y f(p+2) …y f(p+N) ] (4)

[0022] Calculate the covariance matrix Σ of past and future observations pp ,Σ ff and the cross-covariance matrix Σ fp , calculated using the following formula:

[0023]

[0024]

[0025]

[0026] Furthermore, in step S3, the Hankel matrix is ​​subjected to singular value decomposition to find the maximum correlation between past and future observations, as shown in the following formula.

[0027]

[0028] Based on past observations, the projection matrices J and L are calculated to obtain the reduced-dimensional canonical variate matrix X est , the formula is as follows.

[0029]

[0030] X est =JY P (10)

[0031] Further, in step S4, the canonical variable matrix X est As the input data of the subsequent model, it is divided into training data p1, t1 and test data p2, t2, and data normalization is performed, where the input data P = [p1p2] and the output data T = [t1t2].

[0032] [p1,minp,maxp,t1,mint,maxt]=premnmx(P,T) (11)

[0033] Furthermore, in step S5, the network parameters of the extreme learning machine are initialized, the number of neurons in the hidden layer of the ELM is set, the optimization target is set, the population and the optimization parameters are initialized, the fitness function of the GA algorithm is set for the neural network training error, and the fitness function is calculated through operations such as selection, crossover, and mutation. After the optimization target is met, the optimized parameters are used as the weight values ​​and bias values ​​between the input layer and the hidden layer of the ELM to improve the generalization and prediction capabilities of the model.

[0034] [w1,b1,w2,b2,val]=GA(x) (12)

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The chiller load prediction method provided by the present invention has flexibility in processing the dynamic characteristics of data, optimizes energy utilization, improves data processing capabilities, reduces the impact of data sequence correlation through standard variable analysis, and reflects the correlation and difference between different data features to the greatest extent, which is a profound insight into the complexity of data.

[0037] 2. The present invention utilizes canonical variable analysis to effectively reduce the data dimension, realizes effective dimensionality reduction of multi-dimensional data, and realizes rapid processing of multi-dimensional complex information, which is a further exploration of the boundaries of dimensionality reduction methods.

[0038] 3. The present invention ingeniously combines canonical variable analysis with genetically optimized extreme learning machines to build a load forecasting model, which not only significantly improves the prediction accuracy and enhances the stability of the model, effectively avoids random initialization parameters, and uses genetic algorithms to obtain optimized parameters, thereby improving load forecasting accuracy to cope with the ever-changing load demand; it provides strong technical support for the direction of load forecasting and is expected to be widely used and promoted in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0040] Figure 1 The present invention is a flow chart of optimizing the parameters of the extreme learning machine using the GA genetic algorithm.

[0041] Figure 2 A flow chart of building a CVA-GA-ELM model for the method of the present invention.

[0042] Figure 3 It is the fitness function of the genetic algorithm in Example 1 of the present invention.

[0043] Figure 4 This is a diagram of the prediction results of the CVA-GA-ELM model in Example 1 of the present invention.

[0044] Figure 5 This is a comparison chart of the predictions of the CVA-GA-ELM and PCA-BP models in Example 1 of the present invention.

[0045] Figure 6 It is the fitness function of the genetic algorithm in Example 2 of the present invention.

[0046] Figure 7 This is a diagram of the prediction results of the CVA-GA-ELM model in Example 2 of the present invention.

[0047] Figure 8 This is a comparison chart of the predictions of the CVA-GA-ELM and PCA-BP models in Example 2 of the present invention. DETAILED DESCRIPTION

[0048] 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. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0049] Example 1

[0050] See also Figure 1 and Figure 5 The technical solution provided in this embodiment is a chiller load prediction method based on the fusion of CVA and genetic optimization ELM. The Building Technology Office of the U.S. Department of Energy has opened some chiller data to the public. The Lawrence Berkeley National Laboratory and the Pacific Northwest National Laboratory have jointly studied the optimization of building energy consumption and energy saving. The chiller data in the open source data is shown in Table 1. Based on this, a chiller load prediction method based on the fusion of CVA and genetic optimization ELM in this embodiment is implemented.

[0051] Table 1 List of variables

[0052]

[0053] The real data of the chiller has 8 input variables, only 1 output data, and the number of samples is 130,000. In order to intuitively express the effectiveness of the proposed chiller load prediction method, this embodiment takes part of the sample data as model training test data.

[0054] The specific steps of this embodiment are as follows:

[0055] Step 1: Divide the collected open source chiller data into past data y p(r) and future data f(r) , set the past observation value length p and the future observation value length f, the length value can be set arbitrarily, in this embodiment p = 1, f = 1, set the initial value of the system model order n, in this embodiment n = 2, where r is a type of ordinal number, that is, the first data y in this embodiment (r-1) That is data (1) , and so on.

[0056]

[0057] Step 2: Construct the Hankel matrix Y p ∈R,Y f ∈R, which is defined as follows. Assuming l observed variables, then y p(p+1) The last element is y (1) ,y f(p+N) The last element is y (l) , that is, the maximum number of columns of the Hankel matrix is ​​N=lp-f+1, and in this embodiment N=8.

[0058] Y p =[y p(p+1) y p(p+2) …y p(p+N) ]; Y f=[y f(p+1) y f(p+2) …y f(p+N) ]

[0059] Calculate the covariance matrix Σ of past and future observations pp ,Σ ff and the cross-covariance matrix Σ fp .

[0060]

[0061] Step 3: Perform singular value decomposition on the Hankel matrix to find the maximum correlation between past and future observations, and then calculate the projection matrices J and L based on past observations to obtain the reduced-dimensional canonical variable matrix X est .

[0062]

[0063] Step 4: Convert the canonical variable matrix X est As the input data of the subsequent model, it is divided into training data p1, t1 and test data p2, t2, and the input P = [p1p2] and output T = [t1t2] data are normalized.

[0064] [p1,minp,maxp,t1,mint,maxt]=premnmx(P,T)

[0065] Step 5: Initialize the extreme learning machine network parameters, set the number of neurons in the ELM hidden layer, set the optimization target, initialize the population and optimization parameters, set the GA algorithm fitness function for the neural network training error, calculate the GA algorithm fitness function through selection, crossover, mutation and other operations, and after meeting the optimization target, use the optimized parameters as the weight and bias values ​​between the input layer and hidden layer of the ELM to improve the generalization and prediction capabilities of the model.

[0066] [w1,b1,w2,b2,val]=GA(x)

[0067] Step 6: Calculate the hidden layer output matrix and weight value matrix, input the training data p1 and t1 into the model training, and use the test data p2 and t2 for verification to obtain the CVA-GA-ELM load forecasting model and obtain the load forecasting accuracy within the acceptable error range.

[0068] Step 7: Use the same data to enter the PCA-BP prediction model, obtain the prediction results, compare the errors with those in this embodiment, and draw the final conclusion.

[0069] This embodiment performs simulation according to the above steps, and can obtain the chiller load prediction accuracy of the CVA-GA-ELM model and the comparative PCA-BP model within the acceptable range, as shown in Table 2.

[0070] Table 2 Chiller load prediction model accuracy

[0071]

[0072] As can be seen from Table 2 above, this embodiment proposes a chiller load forecasting method based on the fusion of CVA and optimized ELM, which greatly improves the accuracy of chiller load forecasting and brings new ideas to the field of energy management. Compared with the PCA-BP prediction model, this embodiment uses canonical variable analysis to achieve effective dimensionality reduction of multi-dimensional data, realizes rapid processing of multi-dimensional complex information, and reflects the correlation and difference between different features of the data to the greatest extent. It has created a new way to reduce dimensions, effectively explored the potential value of multiple variables, significantly improved the accuracy of prediction, and achieved good prediction results. In summary, this embodiment proposes a chiller load forecasting method based on the fusion of CVA and genetically optimized ELM, which shows high adaptability and application value in energy load forecasting.

[0073] Example 2

[0074] See also Figures 6 to 8 The technical solution provided in this embodiment is to predict the load of a chiller based on the real chiller data of Suzhou Baokong Company as an example, and the data variables are shown in Table 3. Based on this, a chiller load prediction method based on the fusion of CVA and genetic optimization ELM in this embodiment is implemented.

[0075] Table 3 Variable list

[0076]

[0077] The real data of the chiller has 11 input variables, only 1 output data, and the number of samples is more than 200. In order to intuitively express the effectiveness of the proposed chiller load prediction method, this embodiment uses this sample data as model training test data.

[0078] The specific steps are similar to those in Example 1 and will not be repeated here.

[0079] Based on Example 1, Example 2 is simulated according to the above steps, and the power consumption load prediction accuracy of the cooler group of the CVA-GA-ELM model and the comparative PCA-BP model can be obtained respectively, as shown in Table 4.

[0080] Table 4 Chiller load prediction model accuracy

[0081]

[0082] It can be seen from Table 4 above that the chiller load prediction method proposed in this embodiment has a prediction accuracy improved by more than 20% compared with the PCA-BP prediction model. This embodiment achieves effective dimensionality reduction and rapid processing of multi-dimensional complex information, reflects the correlation and difference between different features of the data to the greatest extent, effectively mines the potential value of multiple variables, and significantly improves the prediction accuracy. In summary, this embodiment proposes a chiller load prediction method based on the fusion of CVA and genetic optimization ELM, which has high accuracy and good application value in chiller load prediction.

[0083] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A chiller load forecasting method based on the fusion of CVA and genetic optimization ELM, characterized in that: The following steps are involved: Step S1: Divide the collected open source chiller data into past data and future data, set the length p of the past observation value and the length f of the future observation value and the given system order n; Step S2: Construct the Hankel matrix and calculate the covariance matrix Σ of past and future observations pp ,Σ ff and the cross-covariance matrix Σ fp ; Step S3: Perform singular value decomposition on the Hankel matrix, calculate the projection matrices J and L, and obtain the reduced-dimensional canonical variable matrix X est ; Step S4: Convert the canonical variable matrix X est As the input data of the subsequent model, it is divided into training data p1, t1 and test data p2, t2, and data normalization is performed; Step S5: Initialize the extreme learning machine network parameters, set the number of neurons in the hidden layer of the ELM, set the optimization target, initialize the population and optimization parameters, calculate the fitness function of the GA algorithm, and after meeting the optimization target, use the optimized parameters as the weight value and bias value between the input layer and the hidden layer of the ELM; Step S6: Calculate the hidden layer output matrix and weight value matrix, input the training data p1 and t1 into the model training, and use the test data p2 and t2 for verification to obtain the CVA-GA-ELM load forecasting model, and obtain the load forecasting accuracy within the error range; Step S7: Use the same data to enter the PCA-BP prediction model, obtain the prediction results, perform error comparison, and draw the final conclusion.

2. The chiller load forecasting method based on the fusion of CVA and genetic optimization ELM according to claim 1 is characterized in that: In step S1, the collected open source chiller data is divided into past data y p(r) and future data f(r) , where r is a first-order number, p and f refer to the lengths of past and future observations, respectively; 3. The chiller load forecasting method based on the fusion of CVA and genetic optimization ELM according to claim 1 is characterized in that: In step S2, the Hankel matrix Y is constructed p ∈R,Y f ∈R, which is defined as follows: the maximum number of columns of the Hankel matrix is ​​N = lp-f+1, where l is the observed variable; AND p =[and p(p+1) and p(p+2) …and p(p+N) ] (3) AND f =[and f(p+1) and f(p+2) …and f(p+N) ] (4) Calculate the covariance matrix Σ of past and future observations pp ,Σ ff and the cross-covariance matrix Σ fp , calculated using the following formula:

4. The chiller load forecasting method based on the fusion of CVA and genetic optimization ELM according to claim 1 is characterized in that: In step S3, the Hankel matrix is ​​subjected to singular value decomposition to find the maximum correlation between past and future observations. The formula is as follows: Based on past observations, the projection matrices J and L are calculated to obtain the reduced-dimensional canonical variate matrix X est , the formula is as follows: X est =YOU P (10).

5. The chiller load forecasting method based on the fusion of CVA and genetic optimization ELM according to claim 1 is characterized in that: In step S4, the canonical variable matrix X est As the input data of the subsequent model, it is divided into training data p1, t1 and test data p2, t2, and data normalization is performed, where the input data P = [p1p2], the output data T = [t1t2], [p1, minp, maxp, t1, mint, maxt] = premnmx(P, T) (11).

6. The chiller load forecasting method based on the fusion of CVA and genetic optimization ELM according to claim 1 is characterized in that: In the step S5, the network parameters of the extreme learning machine are initialized: the weight value w1 and the bias value b1 between the input layer and the hidden layer, the number of neurons in the hidden layer of the ELM is set, the optimization target is set, the population and the optimization parameters are initialized, the GA algorithm fitness function is set for the neural network training error, the fitness function is calculated through selection, crossover, and mutation operations, and after the optimization target is met, the optimized parameters w2 and b2 are used as the weight value and bias value between the input layer and the hidden layer of the ELM to improve the generalization ability and prediction ability of the model; [w1,b1,w2,b2,val]=GA(x) (12).

Images