CN114777191A - Heating system household valve regulating and controlling method based on neural network algorithm - Google Patents

Abstract

The invention relates to a heating system home valve regulating and controlling method based on a neural network algorithm, and belongs to the technical field of heating system intelligent management. The regulation and control method is realized by an upper computer platform of a heating system, a field heating data collector and a household valve installed at a user side, wherein a household valve regulation and control model based on a neural network is established on the upper computer platform, the household valve regulation and control model based on the neural network takes the isolation degree, whether the user is a top floor or a bottom floor, the side-to-household/middle-household condition, the indoor temperature, the building type and the heat dissipation type as input data of the neural network, and the weighted value as output data. The neural network algorithm based heating system user valve regulation and control method provided by the invention uses the neural network algorithm to train enough user valve regulation and control data, so that a mathematical model for user valve regulation and control is obtained, the balance regulation and control period of a two-network system can be greatly shortened, and the balance rate of the two-network system is improved.

Description

Control method of household valve in heating system based on neural network algorithm

technical field

The invention relates to the technical field of intelligent regulation and management of heating systems, in particular to a method for regulating and controlling household valves of heating systems based on neural network algorithms.

Background technique

In order to improve the regulation efficiency of the heating system, there are more and more applications of intelligent regulation means in the modern heating system. The existing technical solutions for household valve regulation mainly include:

1. The control method based on the relatively consistent return water temperature: adjust the return water temperature of all users to be basically the same;

2. The control method based on the relative uniformity of the average temperature of the supply and return water: adjust the average temperature of the supply and return water of all users to be basically the same.

The disadvantage of the above two methods is that there are many influencing factors in the regulation process of the household valve, such as building type (energy saving/non-energy saving), heat dissipation type (floor heating/hanging piece), residential occupancy rate, unit type (edge household/intermediate household) ), the top and bottom floors, and whether the neighbors are heated, etc. For different types of houses or the number of neighbors heating is different, the heat dissipation is different, which will affect the indoor temperature. If the return water temperature or the average temperature of the supply and return water are the same, the indoor temperature is also very different, and the proportion of users whose room temperature does not meet the standard is still very high. That is to say, the two control schemes cannot achieve a true balance.

3. The method of regulation based on indoor temperature: All users must install a room temperature collector to adjust the room temperature of all users to be basically the same.

The disadvantage of this method is that all users have to install a room temperature collector, and the room temperature collector needs to be able to communicate with the household valve. This method cannot prevent some users from artificially damaging or interfering with the temperature collection of the room temperature collector by other means. If the indoor equipment is damaged or the temperature collection is disturbed, it will not be able to accurately control. Moreover, all users need to install a room temperature collector, and the cost will increase accordingly.

SUMMARY OF THE INVENTION

In order to solve the problems in the prior art, the patent of the present invention designs a heating system household valve control method based on a neural network algorithm, which reduces the use cost of the heating automatic control system, shortens the control time, and improves the balance rate of the two network balance control.

The technical scheme adopted in the present invention is: the control method is realized by the upper computer platform of the heating system, the on-site heating data collector and the household valve with the electric actuator installed at the user end, and the upper computer platform is built with a neural-based network regulation model,

The steps of the control method are: (1) first establish a physical model of the building structure of the heating system, including the number of floors, the number of units, the number of households, the building type, and the heat dissipation type; the building type includes whether it is an energy-saving building or a non-energy-saving building, The type of heat dissipation includes whether to hang radiators for heat dissipation or floor heating pipes to dissipate heat;

(2) According to the physical model of the building structure, calculate the isolation degree of each user, whether it is the top/ground floor, side/middle household, expected indoor temperature, building type, and heat dissipation type;

(3) Carry out training based on the neural network algorithm model. The training process includes forward propagation of the signal. The calculation data obtained in step (2) is input to the input layer of the neural network model, processed by the hidden layer, and passed to the output layer. , get the predicted weight value, and forward it to the household valve through the upper computer platform of the heating system;

(4) The upper computer platform collects the actual return water temperature of all household valves through the heating data collector, calculates the average return water temperature of all household valves, and issues adjustment instructions to the household valves, configures the return water temperature, and configures the Adjustment interval and adjustment times. After receiving the configured return water temperature, the household valve calculates the target return water temperature. The target return water temperature is equal to the weighted value of the configured return water temperature. After multiple adjustments, the actual return water temperature of the household valve gradually tends to It is close to the target return water temperature. At this time, the indoor temperature of each user is close to the set target indoor temperature, and the adjustment is completed.

Further, the training process based on the neural network algorithm model of the step (3) also includes a back-propagation process of the error, specifically: if the actual output of the output layer and the expected output have a larger error value, then transfer the error value. In the back-propagation stage, the error back-propagation is to back-propagate the output error layer by layer through the hidden layer to the input layer, and distribute the error to all units of each layer, so as to obtain the error signal of each layer unit, this error signal is As the basis for correcting the weights of each unit, when the error between the actual output and the target output does not meet the preset accuracy requirements, the neural network will continuously adjust the weights and update the network until the error is less than the preset accuracy, and the training ends.

Further, the specific process of the forward propagation of the signal in the step (3) is: according to the historical data recorded in the previous heating season, including parameters such as user indoor temperature, return water temperature, isolation degree, building structure and outdoor temperature, obtain The sample data of the neural network, there are 6 neurons d in the input layer, which are isolation degree d ₁ , top/bottom floor d ₂ , side/middle house d ₃ , indoor temperature d ₄ , building type d ₅ , heat dissipation type d ₆ ; The hidden layer neuron O has 9; the output neuron P has one, which is the weight value. The sample is divided into two parts: the training set and the test set, and the input sample data is recorded as:

d(m)=[d ₁ (m),d ₂ (m),...d _n (m)], where n is the number of samples, and m is the number of training and learning times;

Data initialization:

Initialize the weight Vij between the input layer and the hidden layer and the weight Wjk between the hidden layer and the output layer, as well as the threshold a of the hidden layer and the threshold b of the output layer, given the learning rate and activation function;

Hidden layer input:

Among them, i is the number of neurons in the input layer, i=1, 2..., 6; j is the number of neurons in the hidden layer, j=1, 2..., 9;

Hidden layer output calculation:

According to the input sample, the output of the hidden layer is calculated forward, and the output of the hidden layer is,

Among them, i is the number of neurons in the input layer, i=1, 2..., 6; j is the number of neurons in the hidden layer, j=1, 2..., 9;

Output layer input:

where k is the number of neurons in the output layer, k=1;

Output layer output calculation:

According to the output of the hidden layer, the output of the output layer is further calculated, and the output of the output layer is,

Where k is the number of neurons in the output layer, k=1.

Further, the error signal reverse propagation process of the step (3) is,

Compare the output of the output layer with the expected value to get the error signal, the error signal is,

e _k =T _k -P _k

Error performance indicators:

where T is the expected output and P is the actual output;

The partial derivative of the error function with respect to the output layer input:

The partial derivative of the output layer input to the connection weight between the hidden layer and the output layer:

The partial derivative of the error function with respect to the connection weights between the hidden and output layers:

Partial derivative of the error function with respect to the hidden layer input:

Partial derivative of the hidden layer input to the connection weight between the input layer and the hidden layer:

The partial derivative of the error function with respect to the connection weights between the input layer and the hidden layer:

i is the number of neurons in the input layer, j is the number of neurons in the hidden layer, k is the number of neurons in the output layer, η is the learning rate, i=6, j=9, η is set to 0.01, V _ij , W _jk are the connection weights between the input layer and the hidden layer and between the hidden layer and the output layer, respectively, and f is a continuously derivable Sigmoid function;

The network error is a function of the weights between the layers, so adjusting the weights of each layer can change the error E. The principle of adjusting the weights between the layers is to reduce the error continuously.

Further, the weight update between the hidden layer and the output layer in the error signal backpropagation process of the step (3):

W _jk (m+1)=W _jk (m)+ΔW _jk (m)=W _jk (m)+η·δ ₁ (m)·Oo _j (m)

Weight update between input layer and hidden layer:

V _ij (m+1)=V _ij (m)+ΔV _ij (m)=V _ij (m)+η·δ ₂ (m)·d _n (m)

i is the number of neurons in the input layer, j is the number of neurons in the hidden layer, k is the number of neurons in the output layer, η is the learning rate, i=6, j=9, η is set to 0.01, V _ij , W _jk are the connection weights between the input layer and the hidden layer and between the hidden layer and the output layer, respectively, and f is a continuously derivable Sigmoid function;

Error judgment: judge whether the error reaches the preset accuracy or the upper limit of the number of learning times. If it meets the requirements, the neural network training ends, otherwise it returns to re-learning.

Further, the host computer platform communicates with the on-site heating data collector through the 4G service, and the heating data collector communicates with the household valves of each user terminal through RS485.

Compared with the prior art, the improvement of the neural network algorithm-based heating system household valve control method designed by the patent of the present invention is that the method is based on the calculation and control based on the neural network algorithm model, and collects enough data for analysis, Comprehensive consideration of various factors affecting room temperature, including the degree of isolation of each user, whether it is a top/ground floor, side/middle household, expected indoor temperature, building type, heat dissipation type, etc. The household valve regulation model established by this method is accurate, stable, and has strong anti-interference ability, which can greatly shorten the balance regulation period of the second network system and improve the balance rate of the second network system.

Description of drawings

Figure 1 is a schematic diagram of a neural network model of a heating system household valve control method based on a neural network algorithm.

Figure 2 is a schematic diagram of a control system of a heating system based on a neural network algorithm.

Fig. 3 is the temperature change curve of the household valve regulation process based on the method of return water temperature consistency.

Figure 4 is the temperature change curve of the household valve control process based on the neural network algorithm

Detailed ways

The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The technical solutions in the embodiments of the present invention are clearly and completely described, and the described embodiments are only examples of a part of the invention, rather than all of them. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

As shown in Figures 1 and 2, the patent of the present invention designs an embodiment of a heating system household valve control method based on a neural network algorithm. The data collector and the household valve with electric actuator installed on the user end are realized, and the control model based on neural network is built in the upper computer platform. The host computer platform communicates with the on-site heating data collector through the 4G service, and the heating data collector communicates with the household valves of each client through RS485.

The steps of the method for regulating the household valve of the heating system based on the neural network algorithm disclosed in the patent of the present invention are as follows: (1) First, establish a physical model of the building structure of the heating system, including the number of floors, the number of units, the number of households, the type of building, and the type of heat dissipation. ;

(2) According to the physical model of the building structure, calculate the isolation degree of each user, whether it is the top/ground floor, side/middle household, expected indoor temperature, building type, and heat dissipation type;

(3) Carry out training based on the neural network algorithm model. The training process includes forward propagation of the signal and back propagation of the error signal. During forward propagation, the calculation data obtained in step (2) is input to the input of the neural network model. Layer, processed by the hidden layer, passed to the output layer, the predicted weight value is obtained, and sent to the upper computer platform of the heating system. If the actual output of the output layer has a large error value with the expected output, it will turn to the back-propagation stage of the error. Error back propagation is to pass the output error back to the input layer layer by layer through the hidden layer, and apportion the error to all units of each layer, so as to obtain the error signal of each layer unit, and this error signal is used as the weight value of each unit to be corrected. basis. When the error between the actual output and the target output does not meet the preset accuracy requirements, the neural network will continuously adjust the weights and update the network until the error is less than the preset accuracy, and the training ends;

(4) The upper computer platform collects the actual return water temperature of all household valves through the heating data collector, calculates the average return water temperature of all household valves, and issues adjustment instructions to the household valves, configures the return water temperature, and configures the Adjustment interval and adjustment times. After receiving the configured return water temperature, the household valve calculates the target return water temperature. The target return water temperature is equal to the weighted value of the configured return water temperature. After multiple adjustments, the actual return water temperature of the household valve gradually tends to It is close to the target return water temperature. At this time, the indoor temperature of each user is close to the set target indoor temperature, and the adjustment is completed.

As shown in Figure 1, the specific process of the forward propagation of the signal in step (3) is: according to the historical data recorded in the previous heating season, including parameters such as the user's indoor temperature, return water temperature, isolation degree, building structure, and outdoor temperature, The sample data of the neural network is obtained. There are 6 neurons d in the input layer, which are isolation degree d ₁ , top/bottom floor d ₂ , side/middle house d ₃ , indoor temperature d ₄ , building type d ₅ , heat dissipation type d ₆ ; There are 9 neurons O in the hidden layer; there is one output neuron P, which is the weight value. The sample is divided into two parts: the training set and the test set, and the input sample data is recorded as:

d(m)=[d ₁ (m),d ₂ (m),...d _n (m)], where n is the number of samples, and m is the number of training and learning times;

Data initialization:

Initialize the weight V _ij between the input layer and the hidden layer and the weight W _jk between the hidden layer and the output layer, as well as the threshold a of the hidden layer and the threshold b of the output layer, given the learning rate and activation function ;

Hidden layer input:

Among them, i is the number of neurons in the input layer, i=1, 2..., 6; j is the number of neurons in the hidden layer, j=1, 2..., 9;

Hidden layer output calculation:

According to the input sample, the output of the hidden layer is calculated forward, and the output of the hidden layer is,

Among them, i is the number of neurons in the input layer, i=1, 2..., 6; j is the number of neurons in the hidden layer, j=1, 2..., 9;

Output layer input:

where k is the number of neurons in the output layer, k=1;

Output layer output calculation:

According to the output of the hidden layer, the output of the output layer is further calculated, and the output of the output layer is,

Where k is the number of neurons in the output layer, k=1.

Further, the error signal reverse propagation process of the step (3) is,

Compare the output of the output layer with the expected value to get the error signal, the error signal is,

e _k =T _k -P _k

Error performance indicators:

where T is the expected output and P is the actual output;

The partial derivative of the error function with respect to the output layer input:

The partial derivative of the output layer input to the connection weight between the hidden layer and the output layer:

The partial derivative of the error function with respect to the connection weights between the hidden and output layers:

Partial derivative of the error function with respect to the hidden layer input:

Partial derivative of the hidden layer input to the connection weight between the input layer and the hidden layer:

The partial derivative of the error function with respect to the connection weights between the input layer and the hidden layer:

i is the number of neurons in the input layer, j is the number of neurons in the hidden layer, k is the number of neurons in the output layer, η is the learning rate, i=6, j=9, η is set to 0.01, V _ij , W _jk are the connection weights between the input layer and the hidden layer and between the hidden layer and the output layer, respectively, and f is a continuously derivable Sigmoid function;

The network error is a function of the weights between the layers, so adjusting the weights of each layer can change the error E. The principle of adjusting the weights between the layers is to reduce the error continuously.

Further, the weight update between the hidden layer and the output layer in the error signal backpropagation process of the step (3):

W _jk (m+1)=W _jk (m)+ΔW _jk (m)=W _jk (m)+η·δ ₁ (m)·Oo _j (m)

Weight update between input layer and hidden layer:

V _ij (m+1)=V _ij (m)+ΔV _ij (m)=V _ij (m)+η·δ ₂ (m)·d _n (m)

i is the number of neurons in the input layer, j is the number of neurons in the hidden layer, k is the number of neurons in the output layer, η is the learning rate, i=6, j=9, η is set to 0.01, V _ij , W _jk are the connection weights between the input layer and the hidden layer and between the hidden layer and the output layer, respectively, and f is a continuously derivable Sigmoid function;

Error judgment: judge whether the error reaches the preset accuracy or the upper limit of the number of learning times. If it meets the requirements, the neural network training ends, otherwise it returns to re-learning.

Example 1

The same building is selected in a community in Tai'an City, Shandong Province. This building is an energy-saving building, all of which are heated by floor heating pipes, with a floor height of 10 floors. There are intermediate units on each floor. The household valve control method based on the return water temperature consistency method is compared with the household valve control method based on the neural network algorithm disclosed in the present invention. The control is carried out according to the target room temperature of 22°C, and the user's indoor temperature is continuously recorded for 9 days from the first day. .

The east household adopts the traditional household valve control method based on the return water temperature consistency method for automatic control. The room temperature results are shown in Table 1, and the temperature changes during the control process are shown in Figure 3.

Table 1

Xihu adopts the household valve control method based on the neural network algorithm disclosed in the present invention for automatic control, and the room temperature results are recorded as shown in Table 2, and the temperature changes during the control process are shown in Figure 4.

Table 2

From the results recorded in Table 1 and Table 2, it can be seen that the results measured and recorded by the traditional control method based on the return water temperature consistency method in the east households show that the temperature of each household begins to stabilize from the 6th day, but even if After stabilization, the temperature of the same household still fluctuates to a certain extent, and the indoor temperature difference between households on different floors is even greater, especially the temperature difference between the top and bottom floor households and the middle households has reached 4 degrees.

The results measured and recorded by the control method based on the neural network algorithm disclosed in the present invention in the west household show that the temperature of each household begins to stabilize from the third day, and the temperature of the same household is basically the same after stabilization, and there is no significant change; The indoor temperature difference between households on different floors is not much different, even the temperature difference between the top-floor households and the middle households is only 1 degree.

To sum up, the neural network algorithm-based household valve control method of the heating system disclosed in the patent of the present invention is accurate, stable, and has strong anti-interference ability for the indoor temperature control of each household. Equilibrium rate of the system.

The above content is only a preferred embodiment of the present invention, and cannot limit the scope of the present invention. All still belong to the scope covered by the present invention.

Claims (6)

1. the household valve control method of heating system based on neural network algorithm, it is characterized in that, described control method is realized by the host computer platform of heating system, on-site heating data collector and the household valve with electric actuator installed on the user side, The host computer platform has established a control model based on neural network algorithm, (1) First establish the physical model of the building structure of the heating system, including the number of floors, the number of units, the number of households, the type of building, and the type of heat dissipation; (2) According to the physical model of the building structure, calculate the isolation degree of each user, whether it is the top/ground floor, side/middle household, expected indoor temperature, building type, and heat dissipation type; (3) Carry out training based on the neural network algorithm model. The training process includes forward propagation of the signal. The calculation data obtained in step (2) is input to the input layer of the neural network model, processed by the hidden layer, and passed to the output layer. , get the predicted weight value, and forward it to the household valve through the upper computer platform of the heating system; (4) The upper computer platform collects the actual return water temperature of all household valves through the heating data collector, calculates the average return water temperature of all household valves, and issues adjustment instructions to the household valves, configures the return water temperature, and configures the Adjustment interval and adjustment times. After receiving the configured return water temperature, the household valve calculates the target return water temperature. The target return water temperature is equal to the weighted value of the configured return water temperature. After multiple adjustments, the actual return water temperature of the household valve gradually tends to It is close to the target return water temperature. At this time, the indoor temperature of each user is close to the set target indoor temperature, and the adjustment is completed. 2. the heating system valve control method based on neural network algorithm according to claim 1, is characterized in that, the training process based on neural network algorithm model of described step (3) also comprises the back-propagation process of error, concrete It is as follows: if the actual output of the output layer has a large error value with the expected output, it is transferred to the back-propagation stage of the error. The error back-propagation is to back-propagate the output error layer by layer to the input layer through the hidden layer, and transfer the error back to the input layer. The error is apportioned to all units of each layer, so as to obtain the error signal of each layer unit. This error signal is used as the basis for correcting the weight of each unit. When the error between the actual output and the target output does not meet the preset accuracy requirements, The neural network will continuously adjust the weights and update the network until the error is less than the preset accuracy, and the training ends. 3. the heating system household valve control method based on neural network algorithm according to claim 2, is characterized in that, the concrete process of the forward propagation of signal in described step (3) is: according to the historical data recorded before heating season , including the user's indoor temperature, return water temperature, isolation degree, building structure and outdoor temperature and other parameters to obtain the sample data of the neural network, there are 6 neurons d in the input layer, which are the isolation degree d ₁ , the top/bottom floor d ₂ , edge/middle house d ₃ , indoor temperature d ₄ , building type d ₅ , heat dissipation type d ₆ ; there are 9 neurons O in the hidden layer; there is one output neuron P, which is a weight value, and the samples are divided into training sets and The test set consists of two parts, and the input sample data is recorded as: d(m)=[d ₁ (m),d ₂ (m),...d _n (m)], where n is the number of samples, and m is the number of training and learning times; Data initialization: Initialize the weight V _ij between the input layer and the hidden layer and the weight W _jk between the hidden layer and the output layer, as well as the threshold a of the hidden layer and the threshold b of the output layer, given the learning rate and activation function ; Hidden layer input:

Among them, i is the number of neurons in the input layer, i=1, 2..., 6; j is the number of neurons in the hidden layer, j=1, 2..., 9; Hidden layer output calculation: According to the input sample, the output of the hidden layer is calculated forward, and the output of the hidden layer is,

Among them, i is the number of neurons in the input layer, i=1, 2..., 6; j is the number of neurons in the hidden layer, j=1, 2..., 9; Output layer input:

where k is the number of neurons in the output layer, k=1; Output layer output calculation: According to the output of the hidden layer, the output of the output layer is further calculated, and the output of the output layer is,

Where k is the number of neurons in the output layer, k=1. 4. the heating system household valve control method based on neural network algorithm according to claim 3, is characterized in that, the error signal back propagation process of described step (3) is, Compare the output of the output layer with the expected value to get the error signal, the error signal is, e _k =T _k -P _k Error performance indicators:

where T is the expected output and P is the actual output; The partial derivative of the error function with respect to the output layer input:

The partial derivative of the output layer input to the connection weight between the hidden layer and the output layer:

The partial derivative of the error function with respect to the connection weights between the hidden and output layers:

Partial derivative of the error function with respect to the hidden layer input:

Partial derivative of the hidden layer input to the connection weight between the input layer and the hidden layer:

The partial derivative of the error function with respect to the connection weights between the input layer and the hidden layer:

i is the number of neurons in the input layer, j is the number of neurons in the hidden layer, k is the number of neurons in the output layer, η is the learning rate, i=6, j=9, η is set to 0.01, V _ij , W _jk are the connection weights between the input layer and the hidden layer and between the hidden layer and the output layer, respectively, and f is a continuously derivable Sigmoid function; The network error is a function of the weights between the layers, so adjusting the weights of each layer can change the error E. The principle of adjusting the weights between the layers is to reduce the error continuously. 5. the heating system valve control method based on neural network algorithm according to claim 4, is characterized in that, in the error signal back-propagation process of described step (3), the weight between hidden layer and output layer renew:

W _jk (m+1)=W _jk (m)+ΔW _jk (m)=W _jk (m)+η·δ ₁ (m)·Oo _j (m) Weight update between input layer and hidden layer:

V _ij (m+1)=V _ij (m)+ΔV _ij (m)=V _ij (m)+η·δ ₂ (m)·d _n (m) i is the number of neurons in the input layer, j is the number of neurons in the hidden layer, k is the number of neurons in the output layer, η is the learning rate, i=6, j=9, η is set to 0.01, V _ij , W _jk are the connection weights between the input layer and the hidden layer and between the hidden layer and the output layer, respectively, and f is a continuously derivable Sigmoid function; Error judgment: judge whether the error reaches the preset accuracy or the upper limit of the number of learning times. If it meets the requirements, the neural network training ends, otherwise it returns to re-learning. 6. the heating system household valve control method based on neural network algorithm according to claim 5, is characterized in that, described host computer platform communicates with the on-site heating data collector through 4G service, and described heating data collector passes through 4G service. RS485 communicates with the household valve of each user end.

Images