Passive House Energy Supply Method Based on Faster R-CNN Region Detection and Multi-Objective Optimization

Abstract

The present invention proposes a passive house energy supply method based on Faster R-CNN region detection and multi-objective optimization. First, through the Faster R-CNN region detection algorithm, it is detected whether it is in the working state. If it is in the non-working state, the multi-objective optimization algorithm is used to set the best maintenance working state of the equipment to reduce the total energy consumption value. In addition, the Elasitic Net algorithm and the Ridge algorithm are also introduced simultaneously, and different fitting functions are given according to the actual working conditions of each passive house. Furthermore, the multi-objective optimization function can be used to propose the optimal maintenance state temperature values for each location and each time period, and set the targeted maintenance working state.

Description

Technical Field

[0001] The present invention relates to the field of passive house energy supply and distribution, and specifically to a passive house energy supply method based on Faster R-CNN region detection and multi-objective optimization. Background Art

[0002] A passive energy-saving house, also translated as a passive house, is an energy-saving building constructed based on passive design. A passive house can adjust the indoor temperature to a suitable level with very little energy consumption, which is very environmentally friendly. Its basic principle is to adopt energy efficiency. Outstanding thermal insulation walls, innovative window and door technologies, efficient building ventilation, and electrical energy conservation are used to reduce energy consumption.

[0003] Currently, the energy consumption of passive houses is mainly in refrigeration or heating equipment. Traditional passive houses turn off the equipment in the non-working state and use thermal insulation walls to automatically maintain the room temperature during the working state. However, the energy required to restart these refrigeration or heating equipment is often greater than the energy required to maintain the room temperature for a certain period of time. This characteristic results in imperfect utilization of the outstanding thermal insulation performance of passive houses. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a passive house energy supply method based on Faster R-CNN region detection and multi-objective optimization. First, through the Faster R-CNN region detection algorithm, it is detected whether it is in the working state. If it is in the non-working state, the multi-objective optimization algorithm is used to set the optimal maintenance working state of the equipment to reduce the total energy consumption value. In addition, the Elastic Net algorithm and the Ridge algorithm are also introduced simultaneously, and different fitting functions are given according to the actual working conditions of each passive house. Furthermore, the multi-objective optimization function can be used to propose the optimal maintenance state temperature values for each location and each time period, and set the targeted maintenance working state.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A passive house energy supply method based on Faster R-CNN region detection and multi-objective optimization, characterized in that the specific steps are as follows:

[0007] Step 1, collect pictures on-site and transmit them to the server;

[0008] Step 2, process the pictures transmitted to the server, convert the format of the collected pictures, and use the Sobel operator, erosion, and dilation operations to mark the human features;

[0009] Step 3: Select the anchor points, anchor boxes, and the ROI Pooling layer for the marked images to form feature maps, and send them to the subsequent network for training and prediction;

[0010] Step 4: The multi-objective optimization algorithm is specifically expressed as:

[0011] Taking the equipment energy maintenance consumption and the energy consumption for restoring the working room temperature as the optimization objectives, construct a multi-objective optimization scheduling objective function; optimize and solve the objective function to obtain the set value of the maintained temperature;

[0012] Among them, the objective function of the energy consumption of the temperature maintenance device in the passive house is:

[0013] W a = T a * P * t a ;

[0014] In the formula, W a is the total consumption in the maintenance state, T a is the temperature coefficient in the maintenance state, P is the rated power of the equipment, and t a is the set time in the maintenance state;

[0015] The objective function of the energy consumption required by the room temperature equipment when returning to the working state is:

[0016] W β = T β * P * t β

[0017] Among them, W β is the energy required by the air conditioner for returning to the working temperature, T β is the difference coefficient between the working temperature and the maintenance state temperature, and t β is the heating time required by the air conditioner for returning to the working temperature;

[0018] Step 5: Use the Flasitic Net regression algorithm to perform fitting calculations on T a , and the loss function is expressed as:

[0019]

[0020] In the formula, J(θ) is the loss function of the weights. Use the gradient descent method to solve the minimum value of this formula to obtain the weight θ vector value. Among them, λ and p belong to hyperparameters, with the range of λ > 0 and 0 <= p <= 1, and y (i) is the label, which has five features x (i) , and the features need to be normalized;

[0021] Step 6: Use the Ridge regression algorithm to perform fitting calculations on T β , and the loss function is expressed as:

[0022]

[0023] Wherein, J(θ) is the loss function of the weight. The gradient descent method is used to solve the minimum value of this formula to obtain the weight θ vector value, where λ belongs to the hyperparameter and the range is λ > 0. y (i) is the label, which has a total of 6 features x (i) , and the features also need to be normalized;

[0024] Step 7, T a The actual acquisition formula is specifically expressed as:

[0025]

[0026] P is the rated power, t a is the maintenance time, in hours, W a is the total energy consumption in the maintained state;

[0027] Step 8, T β The actual acquisition formula is specifically expressed as:

[0028]

[0029] P is the rated power, t β is the maintenance time, in hours, W β is the total energy consumption for returning to the working room temperature.

[0030] As a further improvement of the present invention, the conversion format in the said step 2 includes noise reduction and smoothing operations.

[0031] Beneficial effects:

[0032] 1. The present invention performs low-frequency sampling analysis on the working scenario through Faster R-CNN, reduces the server pressure, has a simple structure, and a fast training speed.

[0033] 2. The Elastic Net regression algorithm and Ridge algorithm of the present invention perform fitting calculations on T a , T β , and the eigenvalue is determined as the test value of the real-time scenario, and the temperature coefficient values can be proposed for each location and each time period, with the characteristics of strong pertinence and high efficiency.

[0034] 3. The present invention uses the multi-objective optimization algorithm, utilizes the fitting functions of T a , T β to select the best maintained state temperature, and greatly reduces the energy consumption in the passive house. Description of the drawings

[0035] Figure 1 This is the flow chart of the present invention;

[0036] Figure 2 This is the key parameter T of the present invention a Obtaining path diagram;

[0037] Figure 3 This is the key parameter T of the present invention β Obtaining path diagram. Specific implementation manners

[0038] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners:

[0039] The present invention proposes a passive house energy supply method based on Faster R-CNN region detection and multi-objective optimization algorithm. The flow chart of the present invention is as Figure 1 shown. The steps of the present invention will be introduced in detail below in conjunction with the flow chart.

[0040] Step 1: Use the acquisition device to sample pictures of the scene at a frame rate of 0.032, which is only about two samples per minute, reducing the processing pressure on the server.

[0041] Step 2: Conduct preliminary processing on the acquired pictures. First, perform format conversion to convert the picture format to PNG format. And perform gray level equalization, smoothing processing, and Gaussian filtering operations on the image. In addition, for marking the image, first use the Sobel operator to extract edges, then erode and dilate the extracted binary edge image, and finally use the Hough transform for line detection to screen out and mark the human features.

[0042] Step 3: In the RPN network, the corresponding mapping area in the original image is consistent with each point in the feature map, and the anchor is selected at the center of the original image mapping area. According to the human features, four regions with aspect ratios of 2:1, 5:1 and two areas of 256*256, 512*512 are used as anchor boxes, corresponding to the sitting and standing postures of human features, to determine whether there are human features in the scene. Anchor Scaleso is {256×256, 512×512}, and Aspect Ratios is {2:1, 5:1}. The first 50 items in the candidate region proposals generated by the RPN network are sent to the ROI Pooling layer to form a Proposal box on the feature map to obtain the feature map, which is sent to the subsequent network. If there are no human features in 20 consecutive processed pictures, that is, in a 10-minute scene, it is determined that there are no staff in the scene.

[0043] Step 4: Given the device maintenance value through a multi-objective optimization algorithm. First, establish an optimization objective function for the energy consumption of the refrigeration and heating device. The objective function includes: the objective function for minimizing the energy consumption of the temperature maintenance device in the passive house, and the objective function for minimizing the energy consumption required by the room temperature device when returning to the working state.

[0044] The objective function for the energy consumption of the temperature maintenance device in the passive house is:

[0045] W a = T a * P * t a (1)

[0046] In the formula, W a is the total consumption in the maintenance state, and T a is the temperature coefficient in the maintenance state, which is given according to the state in the room at each moment. P is the rated power of the device, and t a is the set time in the maintenance state.

[0047] The objective function for the energy consumption required by the room temperature device when returning to the working state is:

[0048] W β = T β * P * t β (2)

[0049] Where W β is the energy required by the air conditioner at the return working temperature, and T β is the difference coefficient between the working temperature and the maintenance state temperature, which is given according to the state change in the room at each moment, and t β is the heating time required by the air conditioner at the return working temperature.

[0050] For the t β in this objective function,

[0051] t β < 1.5 (3)

[0052] Step 5: Use the Elastic Net regression algorithm to perform fitting calculation on T a The loss function J(θ) is:

[0053]

[0054] In the formula, J(θ) is the loss function of the weight. Use the gradient descent method to solve the minimum value of this formula to obtain the weight θ vector value. Among them, λ and p belong to hyperparameters, and the range is λ > 0, 0 <= p <= 1. 标 Let y (i) be the temperature coefficient T a . The feature x (i)They are respectively: X1 is always 1 and belongs to the extended term, X2 is the set value of the maintenance temperature in the non-working state, X3 is the heat conduction performance of the passive house wall, X4 is the external temperature, and X5 is the volume of the passive house room. Among them, the features need to be normalized.

[0055] The eigenvalue of X2 has the following constraints:

[0056] X2 > X4 (equipment heating state)

[0057] X2 < X4 (equipment cooling state)

[0058] Step 6: Use the Ridge regression algorithm to perform regression fitting on T β The loss function J(θ) is:

[0059]

[0060] In the formula, J(θ) is the loss function of the weights. Use the gradient descent method to solve the minimum value of this formula to obtain the weight θ vector value. Among them, λ belongs to the hyperparameter, and the range is λ > 0. The label y (i) is the difference coefficient T between the working temperature and the maintained state temperature β . The feature x (i) They are respectively: X1 is always 1 and belongs to the extended term, X2 is the set value of the maintenance temperature in the non-working state, X3 is the heat conduction performance of the passive house wall, X4 is the external temperature of the passive house, X5 is the volume of the passive house room, and X6 is the working state temperature value. Among them, the features need to be normalized.

[0061] The eigenvalue of X2 has the following constraints:

[0062] X2 > X4 (equipment heating state)

[0063] X2 < X4 (equipment cooling state)

[0064] Step 7 is as Figure 2 , T a The actual collected data is based on the actual application scenario. Record the total energy consumption W maintained under the rated working efficiency P at a certain time t when different X1, X2, X3, and X4 a are present. a Obtain the values of the label T under different eigenvalue combinations: a

[0065]

[0066] Furthermore, complete the sampling of the training set and the test set.

[0067] Step 8 is as Figure 3 , T βDuring the actual data collection process, in an actual application scenario, record the total energy consumption W maintained under a rated working efficiency P at a certain time t when different values of X1, X2, X3, X4, X5, and X6 are involved. β to obtain the values of the label T under different eigenvalue conditions. β β are as follows:

[0068]

[0069] As described above, it is only a preferred embodiment of the present invention and does not impose any other form of limitation on the present invention. Any modification or equivalent change made based on the technical essence of the present invention still falls within the scope claimed by the present invention.

Claims

1. A passive house energy supply method based on Faster R-CNN region detection and multi-objective optimization, characterized in that, The specific steps are as follows: Step 1: Collect pictures on site and upload them to the server; Step 2: Process the pictures uploaded to the server, convert the format of the collected pictures, and use Sobel operators, erosion, and dilation operations to mark personnel features; Step 3: Select anchor points, anchor boxes, and an ROI Pooling layer for the marked pictures to form a feature map, and send it to the subsequent network for training and prediction; Step 4: The multi-objective optimization algorithm is specifically expressed as: Taking the equipment energy maintenance consumption and the energy consumption for returning to the working room temperature as the optimization objectives, construct a multi-objective optimization scheduling objective function; optimize and solve the objective function to obtain the set value of the maintained temperature; Among them, the objective function of the energy consumption of the temperature maintenance device in the passive house for maintaining the temperature is: W a = T a * P * t a ; Where W a is the total consumption in the maintenance state, T a is the temperature coefficient in the maintenance state, P is the rated power of the device, and t a is the set time in the maintenance state; The objective function of the energy consumption required by the room temperature equipment when returning to the working state is: W β = T β * P * t β Among which W β is the energy required for the air conditioner to return to the working temperature, T β is the difference coefficient between the working temperature and the maintained state temperature, t β is the time required for the air conditioner to heat up to return to the working temperature; Step 5, use the Elasitic Net regression algorithm to perform fitting calculation on T a The loss function is expressed as: Where J(θ) is the loss function of the weights. The gradient descent method is used to solve the minimum value of this formula to obtain the weight θ vector value. Among them, λ and p belong to hyperparameters, and the ranges are λ > 0, 0 <= p <= 1, and y (i) is the label with five features x (i) , and the features need to be normalized; Step 6, use the Ridge regression algorithm to fit T β and the loss function is expressed as: Wherein, J(θ) is the loss function of the weights. The gradient descent method is used to solve the minimum value of this formula to obtain the vector value of the weights θ. Among them, λ belongs to the hyperparameter, and the range is λ > 0, and y (i) is the label, which has a total of 6 features x (i) , and the features also need to be normalized; Step 7, T a The actual acquisition formula is specifically expressed as: P is the rated power, T a is the maintenance time, in hours, W a is the total energy consumption in the maintained state; Step 8, T β The actual acquisition formula is specifically expressed as: P is the rated power, t β is the holding time, in hours, W β is the total energy consumption to return to the studio temperature.

2. The passive house energy supply method based on Faster R-CNN region detection and multi-objective optimization according to claim 1, characterized in that The conversion format in Step 2 includes noise reduction and smoothing operations.

Images