Energy saving prediction method and device for ice storage air conditioning

Abstract

The present application relates to ice storage air conditioning energy saving technical field, particularly point to a kind of ice storage air conditioning energy saving prediction method and device, the method comprises: S1, the index parameter of input training sample's ice storage air conditioning operation time to energy saving prediction model, the index parameter includes as follows any two kinds: temperature when refrigeration machine refrigeration, refrigeration machine condensing pressure, ice volume, ice capacity constraint, system cold balance rate, electric balance weight coefficient, air conditioning cold and heat load coefficient, compressor refrigeration power, between refrigerant and ice water mixture Heat exchange efficiency;S2, the energy saving prediction model is trained by adjusting the weight of the energy saving prediction model, until the construction of the energy saving prediction model is completed;S3, the energy saving situation of the ice storage air conditioning to be predicted is predicted using the energy saving prediction model built.The present application can be well predicted to the energy saving situation of the ice storage air conditioning to be predicted.

Description

Technical Field

[0001] This invention relates to the field of ice storage air conditioning energy-saving technology, and in particular to an ice storage air conditioning energy-saving prediction method and device. Background Technology

[0002] Ice storage air conditioning utilizes ice to store thermal energy. This process can reduce the energy used for cooling during peak electricity demand periods. Alternative energy sources such as solar power can also use this technology to store energy for later use. This is practical because water has a high heat of fusion. While there is considerable research on energy-saving control for ice storage air conditioning, there is currently a lack of research on how to effectively predict its energy-saving performance. Summary of the Invention

[0003] This invention provides a method and apparatus for predicting the energy-saving performance of ice storage air conditioners, enabling accurate prediction of the energy-saving potential of such air conditioners. The technical solution is as follows:

[0004] On the one hand, a method for predicting energy savings in ice storage air conditioning is provided, the method comprising:

[0005] S1. Input the index parameters of the ice storage air conditioner during operation of the training sample into the energy saving prediction model. The index parameters include any two of the following: the temperature when the refrigeration unit is cooling, the condensing pressure of the refrigeration unit, the ice storage volume, the ice storage capacity constraint, the system cold balance rate, the power balance weight coefficient, the air conditioning cooling and heating load coefficient, the compressor cooling power, and the heat exchange efficiency between the refrigerant and the ice-water mixture.

[0006] S2. Train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model until the construction of the energy-saving prediction model is completed;

[0007] S3. Use the constructed energy-saving prediction model to predict the energy-saving performance of the ice storage air conditioner to be predicted.

[0008] Optionally, the energy-saving prediction model includes a BP neural network model, which includes an input layer, a hidden layer, and an output layer. Two index parameters, x1 and x2, are input into the input layer, the inputs y1 and y2 of the hidden layer are calculated, and finally the output z of the output layer is calculated. The output z represents the output factor that minimizes the environmental impact, including the lowest power consumption or the lowest carbon dioxide emissions.

[0009] With x 1(n) ,x 2(n) Let and represent the input data for the nth training iteration; then

[0010] y 1(n) =f(x) 1(n) *W 1(n) +x2(n) *W 3(n) +W 5(n) )

[0011] y 2(n) =f(x) 1(n) *W 2(n) +x 2(n) *W 4(n) +W 6(n) )

[0012] The activation function of the hidden layer neurons in the above formula is: f(x)=-1+max(0,x)+x·tanh(e x );

[0013] y 1(n) ,y 2(n) W represents the data corresponding to the hidden layer after the nth training iteration. 1(n) W 2(n) W 3(n) W 4(n) W 5(n) W 6(n) This represents the weights of each neural network after the nth training iteration;

[0014] z (n) =f(y 1(n) *V 1(n) +y 2(n) *V 2(n) +V 3(n) )

[0015] z (n) V represents the output result after the nth training iteration; 1(n) V 2(n) V 3(n) These represent the weights of the nth training iteration from the hidden layer to the output layer after the hidden layer is activated.

[0016] Optionally, the method further includes initializing the weights of the energy-saving prediction model before S1, setting the initial value of the weights to W. 1(0) ∈[-0.35,-0.18], W 2(0) ∈[0.224,0.565], W 3(0) ∈[-0.157,0.008], W 4(0) ∈[0.697,0.898], W 5(0) ∈[-0.757,0.458], W 6(0) ∈[0.057,0.228].

[0017] Optionally, adjusting the weights of the energy-saving prediction model in step S2 specifically includes:

[0018]

[0019]

[0020]

[0021]

[0022]

[0023]

[0024] In the above formulas, W 1(n+1) W 2(n+1) W 3(n+1) W 4(n+1) W 5(n+1) W 6(n+1) ω represents the function that determines the weights of each neural network after the (n+1)th training iteration from the input layer to the hidden layer; ω represents the learning rate.

[0025] δ Z(n) For correction functions, d (n) This represents the expected output result after the nth training iteration.

[0026] Optionally, the learning rate ω can be set to a value less than 2.75 during training, up to a value of 0.25 where the prediction accuracy is highest.

[0027] On the other hand, an ice storage air conditioning energy-saving prediction device is provided, the device comprising:

[0028] The input module is used to input the index parameters of the ice storage air conditioner during the operation of the training sample into the energy-saving prediction model. The index parameters include any two of the following: the temperature when the refrigeration unit is cooling, the condensing pressure of the refrigeration unit, the ice storage volume, the ice storage capacity constraint, the system cold balance rate, the power balance weight coefficient, the air conditioning cooling and heating load coefficient, the compressor cooling power, and the heat exchange efficiency between the refrigerant and the ice-water mixture.

[0029] The training module is used to train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model until the construction of the energy-saving prediction model is completed;

[0030] The prediction module is used to predict the energy-saving performance of the ice storage air conditioner using a pre-built energy-saving prediction model.

[0031] Optionally, the energy-saving prediction model includes a BP neural network model, which includes an input layer, a hidden layer, and an output layer. Two index parameters, x1 and x2, are input into the input layer, the inputs y1 and y2 of the hidden layer are calculated, and finally the output z of the output layer is calculated. The output z represents the output factor that minimizes the environmental impact, including the lowest power consumption or the lowest carbon dioxide emissions.

[0032] With x 1(n) ,x 2(n) Let and represent the input data for the nth training iteration; then

[0033] y 1(n) =f(x) 1(n) *W 1(n) +x 2(n) *W 3(n) +W 5(n) )

[0034] y 2(n) =f(x) 1(n) *W 2(n) +x 2(n) *W 4(n) +W 6(n) )

[0035] The activation function of the hidden layer neurons in the above formula is: f(x)=-1+max(0,x)+x·tanh(e x );

[0036] y 1(n) ,y 2(n) W represents the data corresponding to the hidden layer after the nth training iteration. 1(n) W 2(n) W 3(n) W 4(n) W 5(n) W 6(n) This represents the weights of each neural network after the nth training iteration;

[0037] z (n) =f(y 1(n) *V 1(n) +y 2(n) *V 2(n) +V 3(n) )

[0038] z (n) V represents the output result after the nth training iteration; 1(n) V 2(n) V 3(n) These represent the weights of the nth training iteration from the hidden layer to the output layer after the hidden layer is activated.

[0039] Optionally, the device further includes an initialization module for initializing the weights of the energy-saving prediction model, setting the initial value of the weights to W. 1(0) ∈[-0.35,-0.18], W 2(0) ∈[0.224,0.565], W 3(0) ∈[-0.157,0.008], W 4(0) ∈[0.697,0.898], W 5(0) ∈[-0.757,0.458], W 6(0) ∈[0.057,0.228].

[0040] Optionally, the training module is used to train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model, specifically including:

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047] In the above formulas, W 1(n+1) W 2(n+1) W 3(n+1) W 4(n+1) W 5(n+1) W 6(n+1) ω represents the function that determines the weights of each neural network after the (n+1)th training iteration from the input layer to the hidden layer; ω represents the learning rate.

[0048] δ Z(n) For correction functions, d (n) This represents the expected output result after the nth training iteration.

[0049] Optionally, the training module can use a learning rate ω that starts from less than 2.75 during training, until the prediction accuracy is highest when ω is 0.25.

[0050] On the other hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described ice storage air conditioner energy-saving prediction method.

[0051] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-described ice storage air conditioner energy-saving prediction method.

[0052] The beneficial effects of the technical solution provided by this invention include at least the following:

[0053] The present invention provides a method and apparatus for predicting energy-saving performance of ice storage air conditioning, which can accurately predict the energy-saving performance of ice storage air conditioning, thereby achieving energy-saving and emission-reduction benefits such as carbon peaking and carbon neutrality. The aim is to use the simplest neural network to explore the most accurate prediction precision, so as to make the best decision for users to make further plans. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of an ice storage air conditioner energy-saving prediction method provided by an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of a BP neural network provided in an embodiment of the present invention;

[0057] Figure 3 This is a detailed flowchart of an ice storage air conditioning energy-saving prediction method provided by an embodiment of the present invention;

[0058] Figure 4 This is a diagram illustrating the process of adjusting the initial value of W1 according to an embodiment of the present invention;

[0059] Figure 5 This is a block diagram of an ice storage air conditioner energy-saving prediction device provided in an embodiment of the present invention;

[0060] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0061] like Figure 1 As shown in the figure, an embodiment of the present invention provides an energy-saving prediction method for ice storage air conditioning, the method comprising:

[0062] S1. Input the index parameters of the ice storage air conditioner during operation of the training sample into the energy saving prediction model. The index parameters include any two of the following: the temperature when the refrigeration unit is cooling, the condensing pressure of the refrigeration unit, the ice storage volume, the ice storage capacity constraint, the system cold balance rate, the power balance weight coefficient, the air conditioning cooling and heating load coefficient, the compressor cooling power, and the heat exchange efficiency between the refrigerant and the ice-water mixture.

[0063] S2. Train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model until the construction of the energy-saving prediction model is completed;

[0064] S3. Use the constructed energy-saving prediction model to predict the energy-saving performance of the ice storage air conditioner to be predicted.

[0065] The following is combined Figures 2-4 The present invention provides a detailed description of the energy-saving prediction method for ice storage air conditioning, the method comprising:

[0066] S1. Input the index parameters of the ice storage air conditioner during operation of the training sample into the energy saving prediction model. The index parameters include any two of the following: the temperature when the refrigeration unit is cooling, the condensing pressure of the refrigeration unit, the ice storage volume, the ice storage capacity constraint, the system cold balance rate, the power balance weight coefficient, the air conditioning cooling and heating load coefficient, the compressor cooling power, and the heat exchange efficiency between the refrigerant and the ice-water mixture.

[0067] Optionally, such as Figure 2 As shown, the energy-saving prediction model includes a BP neural network model, which includes an input layer, a hidden layer, and an output layer. Two index parameters, x1 and x2, are input into the input layer, the inputs y1 and y2 of the hidden layer are calculated, and finally the output z of the output layer is calculated. The output z represents the output factor that minimizes the environmental impact, including the lowest power consumption or the lowest carbon dioxide emissions.

[0068] With x 1(n) ,x 2(n) Let and represent the input data for the nth training iteration; then

[0069] y 1(n) =f(x) 1(n) *W 1(n) +x 2(n) *W 3(n) +W 5(n) )

[0070] y 2(n) =f(x) 1(n) *W 2(n) +x 2(n) *W 4(n) +W 6(n) )

[0071] The activation function of the hidden layer neurons in the above formula is: f(x)=-1+max(0,x)+x·tanh(e x );

[0072] y 1(n) ,y 2(n) W represents the data corresponding to the hidden layer after the nth training iteration. 1(n) W 2(n) W 3(n) W 4(n) W 5(n) W 6(n) This represents the weights of each neural network after the nth training iteration;

[0073] z (n) =f(y 1(n) *V 1(n) +y 2(n) *V 2(n) +V 3(n) )

[0074] z (n) V represents the output result after the nth training iteration; 1(n) V 2(n) V 3(n) These represent the weights of the nth training iteration from the hidden layer to the output layer after the hidden layer is activated.

[0075] Bias variables b1 = b2 = 1.

[0076] The activation function of the hidden layer neurons is: f(x) = -1 + max(0,x) + x·tanh(e^x) x The function is chosen because it has a good fit. From a machine learning perspective, the fuzzy hyperplane promotes class separation. Therefore, using a non-linear activation function in the training step increases the network's ability to process more complex data.

[0077] The input layer variables are: X = [x1, x2] T

[0078] x1 is the first indicator parameter, and x2 is the second indicator parameter. These two indicator parameters can be any two of the following: the temperature when the refrigeration unit is cooling, the condensing pressure of the refrigeration unit, the ice storage volume, the ice storage capacity constraint, the system cold balance rate, the electrical balance weighting coefficient, the air conditioning cooling and heating load coefficient, the compressor cooling power, and the heat exchange efficiency between the refrigerant and the ice-water mixture.

[0079] definition:

[0080] In the above formula, δ Z(n) For the correction function, d (n) z represents the expected output (actual accurate value) after the nth training iteration.(n) This represents the output result after the nth training iteration. Different initial weight values ​​will lead to different results. Extensive computation shows that the learning rate ω affects the initial value of W1. The process of adjusting the initial value of W1 is as follows... Figure 3 As shown, the rest are similar.

[0081] Optionally, such as Figure 4 As shown, the method further includes initializing the weights of the energy-saving prediction model before S1, setting the initial weight values ​​to W. 1(0) ∈[-0.35,-0.18], W 2(0) ∈[0.224,0.565], W 3(0) ∈[-0.157,0.008], W 4(0) ∈[0.697,0.898], W 5(0) ∈[-0.757,0.458], W 6(0) ∈[0.057,0.228], which can prevent the weight from being far away from the rejection point.

[0082] S2. Train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model until the construction of the energy-saving prediction model is completed;

[0083] Optionally, adjusting the weights of the energy-saving prediction model in step S2 specifically includes:

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090] In the above formulas, W 1(n+1) W 2(n+1) W 3(n+1) W 4(n+1) W 5(n+1) W 6(n+1) ω represents the function that determines the weights of each neural network after the (n+1)th training iteration from the input layer to the hidden layer; ω represents the learning rate.

[0091] δ Z(n) For correction functions, d (n) This represents the expected output result after the nth training iteration.

[0092] Optionally, the learning rate ω can be set to a value less than 2.75 during training, up to a value of 0.25 where the prediction accuracy is highest.

[0093] Extensive experiments have shown that the highest prediction accuracy is achieved when ω is set to 0.25. While this method can be adjusted to common parameters such as 0.3, 0.1, 0.01, and 0.001, experiments have shown that when ω ≥ 2.75, the weight values ​​begin to approach infinity, negating the purpose of machine learning training. Therefore, the highest prediction accuracy is achieved when the learning rate is kept as low as possible below 2.75, ideally down to ω of 0.25.

[0094] Definition: Loss function

[0095] The purpose of the loss function is to determine the difference between the expected value (the actual accurate value) and the output value of the neural network. i (i = 1, 2, ..., n) is the expected value, z i (i = 1, 2, ..., n) is the output of the neural network.

[0096] Figure 4 The "Whether the accuracy requirement is met" setting can be set according to the user's actual needs. Generally, according to the MSE formula, the smaller the MSE exponent, the more accurate the result.

[0097] Some existing technologies propose using t input variables (t≥5) simultaneously to study their impact on the prediction target, which is reasonable. Based on the principle of this invention, the time series of weights can also be used to predict the simultaneous effect of multiple variable factors. However, for multiple variables, it is not conducive to extracting the weight map as described in this invention, as this greatly increases the extraction difficulty. After testing, the computer's accuracy and efficiency in calculating weights decreased significantly, and after multiple training sessions, the correlation between data became smaller and smaller, even less accurate than training individually and precisely adjusting parameters each time to predict the impact on the target. Experiments showed that when using t input variables (t≥5) simultaneously, adjusting the weights each time can easily lead to the weights becoming infinitely large, even reducing the overall prediction accuracy by 6.4%. Therefore, the prediction method of this invention can accurately predict the energy-saving situation of ice storage air conditioning, thereby achieving energy-saving and emission-reduction benefits such as carbon peaking and carbon neutrality. The aim is to use the simplest neural network to explore the most accurate prediction accuracy, providing users with optimal decision-making for further planning.

[0098] Furthermore, regarding the limitation on the number of iterations (training iterations) n, parameter tuning revealed that n ≥ 200 training iterations are required to achieve a relatively clear recursive time series relationship in the weights. However, the value should not be too large (n ≥ 10).5 Otherwise, it will affect the computational efficiency.

[0099] S3. Use the constructed energy-saving prediction model to predict the energy-saving performance of the ice storage air conditioner to be predicted.

[0100] Once the energy-saving prediction model is constructed, the energy-saving performance of the ice storage air conditioner can be predicted. The specific method is as follows:

[0101] Input any two index parameters to be predicted when the ice storage air conditioner is operating into the energy-saving prediction model. The index parameters include any two of the following: temperature when the refrigeration unit is cooling, condensing pressure of the refrigeration unit, ice storage volume, ice storage capacity constraint, system cold balance rate, power balance weight coefficient, air conditioning cooling and heating load coefficient, compressor cooling power, and heat exchange efficiency between refrigerant and ice-water mixture.

[0102] The energy-saving prediction model outputs the energy-saving performance of the ice storage air conditioner to be predicted. Specifically, the energy-saving performance represents the output factor that minimizes environmental impact, including the lowest power consumption or the lowest carbon dioxide emissions. That is, the energy-saving prediction model outputs the lowest power consumption or the lowest carbon dioxide emissions of the ice storage air conditioner to be predicted.

[0103] like Figure 5 As shown in the figure, this embodiment of the invention also provides an ice storage air conditioner energy-saving prediction device, the device comprising:

[0104] The input module 510 is used to input the index parameters of the ice storage air conditioner during the operation of the training sample into the energy saving prediction model. The index parameters include any two of the following: the temperature when the refrigeration unit is cooling, the condensing pressure of the refrigeration unit, the ice storage volume, the ice storage capacity constraint, the system cold balance rate, the power balance weight coefficient, the air conditioning cooling and heating load coefficient, the compressor cooling power, and the heat exchange efficiency between the refrigerant and the ice-water mixture.

[0105] Training module 520 is used to train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model until the construction of the energy-saving prediction model is completed;

[0106] The prediction module 530 is used to predict the energy saving of the ice storage air conditioner to be predicted using the constructed energy saving prediction model.

[0107] Optionally, the energy-saving prediction model includes a BP neural network model, which includes an input layer, a hidden layer, and an output layer. Two index parameters, x1 and x2, are input into the input layer, the inputs y1 and y2 of the hidden layer are calculated, and finally the output z of the output layer is calculated. The output z represents the output factor that minimizes the environmental impact, including the lowest power consumption or the lowest carbon dioxide emissions.

[0108] With x1(n) ,x 2(n) Let and represent the input data for the nth training iteration; then

[0109] y 1(n) =f(x) 1(n) *W 1(n) +x 2(n) *W 3(n) +W 5(n) )

[0110] y 2(n) =f(x) 1(n) *W 2(n) +x 2(n) *W 4(n) +W 6(n) )

[0111] The activation function of the hidden layer neurons in the above formula is: f(x)=-1+max(0,x)+x·tanh(e x );

[0112] y 1(n) ,y 2(n) W represents the data corresponding to the hidden layer after the nth training iteration. 1(n) W 2(n) W 3(n) W 4(n) W 5(n) W 6(n) This represents the weights of each neural network after the nth training iteration;

[0113] z (n) =f(y 1(n) *V 1(n) +y 2(n) *V 2(n) +V 3(n) )

[0114] z (n) V represents the output result after the nth training iteration; 1(n) V 2(n) V 3(n) These represent the weights of the nth training iteration from the hidden layer to the output layer after the hidden layer is activated.

[0115] Optionally, the device further includes an initialization module for initializing the weights of the energy-saving prediction model, setting the initial value of the weights to W. 1(0) ∈[-0.35,-0.18], W 2(0) ∈[0.224,0.565], W 3(0) ∈[-0.157,0.008], W 4(0) ∈[0.697,0.898], W 5(0)∈[-0.757,0.458], W 6(0) ∈[0.057,0.228].

[0116] Optionally, the training module is used to train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model, specifically including:

[0117]

[0118]

[0119]

[0120]

[0121]

[0122]

[0123] In the above formulas, W 1(n+1) W 2(n+1) W 3(n+1) W 4(n+1) W 5(n+1) W 6(n+1) ω represents the function that determines the weights of each neural network after the (n+1)th training iteration from the input layer to the hidden layer; ω represents the learning rate; δ Z(n) For correction functions, d (n) This represents the expected output result after the nth training iteration.

[0124] Optionally, the training module can use a learning rate ω that starts from less than 2.75 during training, until the prediction accuracy is highest when ω is 0.25.

[0125] The energy-saving prediction device for ice storage air conditioning provided in this embodiment of the invention has a functional structure that corresponds to the energy-saving prediction method for ice storage air conditioning provided in this embodiment of the invention, and will not be described again here.

[0126] Figure 6 This is a schematic diagram of the structure of an electronic device 600 provided in an embodiment of the present invention. The electronic device 600 may vary considerably due to different configurations or performance. It may include one or more central processing units (CPUs) 601 and one or more memories 602. The memory 602 stores at least one instruction, which is loaded and executed by the processor 601 to implement the steps of the above-mentioned ice storage air conditioning energy-saving prediction method.

[0127] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the above-described ice storage air conditioner energy-saving prediction method. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device.

[0128] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0129] 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 principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting energy savings in ice storage air conditioning systems, characterized in that, The method includes: S1. Input the index parameters of the ice storage air conditioner during operation of the training sample into the energy saving prediction model. The index parameters include any two of the following: the temperature when the refrigeration unit is cooling, the condensing pressure of the refrigeration unit, the ice storage volume, the ice storage capacity constraint, the system cold balance rate, the power balance weight coefficient, the air conditioning cooling and heating load coefficient, the compressor cooling power, and the heat exchange efficiency between the refrigerant and the ice-water mixture. S2. Train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model until the construction of the energy-saving prediction model is completed; S3. Use the constructed energy-saving prediction model to predict the energy-saving performance of the ice storage air conditioner to be predicted; The energy-saving prediction model includes a BP neural network model, which includes an input layer, a hidden layer, and an output layer. Two index parameters, x1 and x2, are input into the input layer. The inputs y1 and y2 of the hidden layer are calculated. Finally, the output z of the output layer is calculated. The output z represents the output factor that minimizes the environmental impact, including the lowest power consumption or the lowest carbon dioxide emissions. With x 1(n) ,x 2(n) Let and represent the input data for the nth training iteration; then y 1(n) =f(x 1(n) *W 1(n) +x 2(n) *W 3(n) +W 5(n) ) y 2(n) =f(x 1(n) *W 2(n) +x 2(n) *W 4(n) +W 6(n) ) The activation function of the hidden layer neurons in the above formula is: f(x)=-1+max(0,x)+x·tanh(e x ); y 1(n) ,y 2(n) W represents the data corresponding to the hidden layer after the nth training iteration. 1(n) W 2(n) W 3(n) W 4(n) W 5(n) W 6(n) This represents the weights of each neural network after the nth training iteration; z (n) =f(y 1(n) *V 1(n) +y 2(n) *V 2(n) +V 3(n) ) z (n) V represents the output result after the nth training iteration; 1(n) V 2(n) V 3(n) These represent the weights of the nth training iteration from the hidden layer to the output layer after the hidden layer is activated.

2. The method according to claim 1, characterized in that, The method further includes initializing the weights of the energy-saving prediction model before S1, setting the initial value of the weights to W. 1(0) ∈[-0.35,-0.18], W 2(0) ∈[0.224,0.565], W 3(0) ∈[-0.157,0.008], W 4(0) ∈[0.697,0.898], W 5(0) ∈[-0.757,0.458], W 6(0) ∈[0.057,0.228].

3. The method according to claim 1, characterized in that, The adjustment of the weights in the energy-saving prediction model in step S2 specifically includes: In the above formulas, W 1(n+1) W 2(n+1) W 3(n+1) W 4(n+1) W 5(n+1) W 6(n+1) ω represents the function that determines the weights of each neural network after the (n+1)th training iteration from the input layer to the hidden layer; ω represents the learning rate. δ Z(n) For correction functions, d (n) This represents the expected output result after the nth training iteration.

4. The method according to claim 3, characterized in that, During training, the learning rate ω starts from less than 2.75 and continues until ω reaches 0.25, at which point the prediction accuracy is highest.

5. An ice storage air conditioning energy-saving prediction device, characterized in that, The device includes: The input module is used to input the index parameters of the ice storage air conditioner during the operation of the training sample into the energy-saving prediction model. The index parameters include any two of the following: the temperature when the refrigeration unit is cooling, the condensing pressure of the refrigeration unit, the ice storage volume, the ice storage capacity constraint, the system cold balance rate, the power balance weight coefficient, the air conditioning cooling and heating load coefficient, the compressor cooling power, and the heat exchange efficiency between the refrigerant and the ice-water mixture. The training module is used to train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model until the construction of the energy-saving prediction model is completed; The prediction module is used to predict the energy-saving performance of the ice storage air conditioner using the constructed energy-saving prediction model. The energy-saving prediction model includes a BP neural network model, which includes an input layer, a hidden layer, and an output layer. Two index parameters, x1 and x2, are input into the input layer. The inputs y1 and y2 of the hidden layer are calculated. Finally, the output z of the output layer is calculated. The output z represents the output factor that minimizes the environmental impact, including the lowest power consumption or the lowest carbon dioxide emissions. With x 1(n) ,x 2(n) Let and represent the input data for the nth training iteration; then y 1(n) =f(x 1(n) *W 1(n) +x 2(n) *W 3(n) +W 5(n) ) y 2(n) =f(x 1(n) *W 2(n) +x 2(n) *W 4(n) +W 6(n) ) The activation function of the hidden layer neurons in the above formula is: f(x)=-1+max(0,x)+x·tanh(e x ); y 1(n) ,y 2(n) W represents the data corresponding to the hidden layer after the nth training iteration. 1(n) W 2(n) W 3(n) W 4(n) W 5(n) W 6(n) This represents the weights of each neural network after the nth training iteration; z (n) =f(y 1(n) *V 1(n) +y 2(n) *V 2(n) +V 3(n) ) z (n) V represents the output result after the nth training iteration; 1(n) V 2(n) V 3(n) These represent the weights of the nth training iteration from the hidden layer to the output layer after the hidden layer is activated.

6. The apparatus according to claim 5, characterized in that, The device further includes an initialization module for initializing the weights of the energy-saving prediction model, setting the initial value of the weights to W. 1(0) ∈[-0.35,-0.18], W 2(0) ∈[0.224,0.565], W 3(0) ∈[-0.157,0.008], W 4(0) ∈[0.697,0.898], W 5(0) ∈[-0.757,0.458], W 6(0) ∈[0.057,0.228].

7. The apparatus according to claim 5, characterized in that, The training module is used to train the energy-saving prediction model by adjusting the weights of the energy-saving prediction model, specifically including: In the above formulas, W 1(n+1) W 2(n+1) W 3(n+1) W 4(n+1) W 5(n+1) W 6(n+1) ω represents the function that determines the weights of each neural network after the (n+1)th training iteration from the input layer to the hidden layer; ω represents the learning rate. δ Z(n) For correction functions, d (n) This represents the expected output result after the nth training iteration.

8. The apparatus according to claim 7, characterized in that, The training module uses a learning rate ω that starts from less than 2.75 during training, and the prediction accuracy is highest when ω is 0.25.

Images