Power distribution room air conditioner load adjusting method based on model prediction

Abstract

The invention discloses a power distribution room air-conditioning load adjusting method based on model prediction, which designs a power distribution station thermodynamic model from several important aspects of space structure, time structure and power grid load demand, and fully considers the influence of the space size and structure of a power distribution station on heat energy transmission in the power distribution station. Meanwhile, the vehicle problem that the power distribution station absorbs the heat energy based on the time period is considered, the influence of the power grid load on the heat loss of the power distribution equipment is also considered, and the influence on the heat energy in the power distribution station is comprehensively and effectively supervised; meanwhile, along with intelligent development of a power grid, power distribution equipment in the power distribution station is gradually increased, the design of the scheme can also consider the situation that the equipment in the power distribution station is increased, space structure characteristics of the power distribution station are adjusted according to the increased power distribution equipment, and body surface temperature monitoring of the power distribution equipment is increased.

Description

Technical Field

[0001] The invention relates to a method for regulating the air-conditioning load in a power distribution room based on model prediction, belonging to the load control technology of a transformer substation. Background Art

[0002] With the rapid development of the economy and the electric power industry, the number of power facilities and distribution stations that are closely related to people's work and life is increasing, which has brought great convenience to people's electricity use. Due to the increasing requirements for the intelligent level of the power dispatching master station system, there are more and more distribution equipment in the distribution station. The increase in distribution equipment has increased the energy consumption of the distribution station on the one hand, and increased the temperature in the distribution station on the other hand; in order to ensure the safe and stable operation of the distribution station, more attention needs to be paid to the air conditioning control of the distribution station, which should not only ensure the safe and stable operation of the distribution equipment, but also ensure the predictability of energy consumption, so as to reduce the line loss and load control of the entire power grid. Therefore, it is increasingly important to predict and control the air conditioning load of the distribution station. Summary of the invention

[0003] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a method for regulating the air-conditioning load in a distribution room based on model prediction, which predicts and regulates the air-conditioning load of a distribution station.

[0004] Technical solution: To achieve the above purpose, the technical solution adopted by the present invention is:

[0005] A method for regulating air conditioning load in a power distribution room based on model prediction comprises the following steps:

[0006] Step 1: Construct a distribution station spatial structure feature extraction module to extract the distribution station spatial structure feature F space ;

[0007] Step 2: Obtain historical data on heat loss and air conditioning load of power distribution equipment, including time characteristics F time , Distribution network net load active demand characteristics F P , distribution station temperature characteristics F tem and air conditioning cooling capacity label Q ture , divide the historical data into training data set and test data set in proportion;

[0008] Step 3: Construct a thermodynamic model of the distribution station based on a neural network; first, use the spatial fusion module to quantify the spatial structure characteristics of the distribution station. space and the distribution station temperature characteristic F tem Perform feature fusion to obtain spatial fusion features By spatial fusion features Characterize the impact of the spatial structure of the distribution room on the air conditioning load; then use the time fusion module to fuzzy the time characteristics F timeand the distribution station temperature characteristic F tem Perform feature fusion to obtain time fusion features Fusion of features through time Characterize the influence of time information on air conditioning load; then use the active demand fusion module to fusion the active demand characteristics of the net load of the distribution network F P and the distribution station temperature characteristic F tem Perform feature fusion to obtain active demand fusion features Fusion characteristics through active demand Characterize the impact of the net active demand of the distribution network on the air conditioning load; finally, the spatial fusion characteristics are adjusted by adaptively adjusting the fusion module Temporal fusion features and active power demand fusion characteristics Perform multi-dimensional fusion to obtain multi-dimensional fusion features Multi-dimensional fusion features Input to the detection head for semantic segmentation to obtain the predicted value Q of air conditioning cooling capacity pre ;

[0009] Step 4: Use the loss function to supervise the thermodynamic model of the distribution station and obtain a trained thermodynamic model of the distribution station;

[0010] Step 5. Input the spatial structure characteristics of the distribution station and the test data set into the trained distribution station thermodynamic model to obtain the predicted value of air conditioning cooling capacity, and adjust the air conditioning load in the distribution room based on the predicted value of air conditioning cooling capacity.

[0011] This case designs the thermodynamic model of the distribution station from several important aspects, including spatial structure, time structure and grid load demand. It fully considers the impact of the size and structure of the distribution station on the transfer of heat energy in the distribution station, and also considers the problem of the distribution station absorbing heat energy based on the time period. It also considers the impact of the grid load on the heat loss of the distribution equipment, thus achieving comprehensive and effective supervision of the impact of heat energy in the distribution station. At the same time, with the development of the intelligent power grid, the distribution equipment in the distribution station is also gradually increasing. The design of this case can also take into account the increase of equipment in the distribution station, adjust the spatial structure characteristics of the distribution station according to the added distribution equipment, and increase the surface temperature monitoring of the distribution equipment.

[0012] Specifically, in Step 1, the distribution station spatial structure feature F space It includes the exterior wall orientation, exterior wall volume, internal volume, effective volume and window-to-wall ratio of the distribution station calculated based on the geographical location and structural dimensions of the distribution station. The effective volume is the space after deducting the equivalent volume of all distribution equipment from the internal volume of the distribution station. The window-to-wall ratio is the ratio of the area of ​​the heat dissipation windows on each exterior wall of the distribution station to the area of ​​the exterior wall to which it belongs.

[0013] Specifically, in Step 2, the time feature F time Including quarterly information, weekly information, rest day information and daily scheduling time information, the distribution station temperature characteristic F tem Including distribution equipment surface temperature series, outdoor temperature, and sunshine information.

[0014] Specifically, in the Step 3, the structures of the spatial fusion module, the temporal fusion module and the active demand fusion module in the thermodynamic model of the distribution station are the same, including a connection layer, a 1×1 convolution, a 3×3 convolution, a layer normalization layer and a Sigmoid activation function. The feature fusion process is expressed as:

[0015]

[0016]

[0017]

[0018] Among them: Cat represents the feature connection operation, represents a convolution with a 1×1 kernel, denotes a convolution with a 3×3 convolution kernel and Norm denotes a layer normalization operation.

[0019] Specifically, in Step 3, the multi-dimensional fusion feature is obtained by the adaptive adjustment fusion module in the thermodynamic model of the distribution station. The steps include:

[0020] Step 31: First, spatial fusion features Temporal fusion features and active power demand fusion characteristics Add element by element to get Then F1 c Input to the channel attention unit and the spatial attention unit at the same time;

[0021] Step 32: In the channel attention unit, first focus on F1 along the channel dimension. c Perform a global average pooling operation Then, a 1×1 convolution is used to reduce the channel dimension, a ReLU activation function is used for nonlinear enhancement, and another 1×1 convolution is used to reduce the channel dimension. It is expressed as:

[0022]

[0023] Step 33: In the spatial attention unit, first focus F1 along the spatial dimension. c Perform a global average pooling operation And the global maximum pooling operation Then connect along the spatial dimension and input into a 7×7 convolution to get It is expressed as:

[0024]

[0025] in: represents a convolution with a 7×7 convolution kernel;

[0026] Step 34. and Add element by element to get Based on F1 c The corresponding channel pair Each channel of F1 is adjusted first along the channel dimension. c and Connect them, then rearrange each channel in an alternating manner through the channel shuffle operation, then pass the group convolution, and finally pass the Sigmoid activation function to obtain the fused weight map R, which is expressed as:

[0027]

[0028] Among them: CS represents the channel shuffle operation, represents a group convolution with a 7×7 convolution kernel;

[0029] Step 34, map the fusion weight R to the spatial fusion feature Temporal fusion features and active power demand fusion characteristics Performing element-wise multiplication, we get and

[0030] Step 34: Add each feature element by element, and then perform point-by-point convolution to obtain multi-dimensional fusion features. It is expressed as:

[0031]

[0032] in: Represents a point-wise convolution with a 1×1 kernel.

[0033] Specifically, in Step 4, the mean absolute error function MAE and the mean absolute percentage error function MAPE are used as loss functions to supervise the thermodynamic model of the distribution station:

[0034]

[0035]

[0036] Where: K represents the number of samples in the training data set.

[0037] Specifically, the distribution room air conditioning load regulation method is executed by a distribution room air conditioning load regulation system, and the distribution room air conditioning load regulation system includes a distribution station spatial structure feature extraction module, a data acquisition module, a distribution station thermodynamic model and a supervision training module;

[0038] The distribution station spatial structure feature extraction module first calculates the external wall orientation, external wall volume, internal volume, effective volume and window-to-wall ratio of the distribution station based on the geographical location and structural dimensions of the distribution station, and then extracts the distribution station spatial structure feature F based on the calculation results. space ;

[0039] The data acquisition module collects the geographical location and structural dimensions of the distribution station and sends them to the distribution station spatial structure feature extraction module; collects quarterly information, weekly information, rest day information and daily scheduling time information, and sends them to the distribution station thermodynamic model; collects the surface temperature sequence of the distribution equipment, outdoor temperature, and sunshine information, and sends them to the distribution station thermodynamic model; collects the net load active power demand of the distribution network, and sends it to the distribution station thermodynamic model;

[0040] The distribution station thermodynamic model includes a distribution station temperature feature extraction module, a space fusion module, a time fusion module, an active demand fusion module, an adaptive adjustment fusion module and a detection head; the distribution station temperature feature extraction module is used to extract the distribution station temperature feature F tem ; The spatial fusion module is used to integrate the spatial structure characteristics F of the distribution station space and the distribution station temperature characteristic F tem Get spatial fusion features The time fusion module is used to extract the time feature F time , and integrate the temporal feature F time and the distribution station temperature characteristic F tem Get time fusion features The active demand fusion module is used to extract the active demand characteristics F of the net load of the distribution network. P , and integrate the active demand characteristics of the distribution network net load F P and the distribution station temperature characteristic F tem Get the active demand fusion feature F P ; The adaptive adjustment fusion module adjusts the spatial fusion features Temporal fusion features and active power demand fusion characteristics Perform multi-dimensional fusion to obtain multi-dimensional fusion features The detection head uses multi-dimensional fusion features Perform semantic segmentation to obtain the predicted value Q of air conditioning cooling capacitypre ;

[0041] The supervised training module uses a loss function to perform supervised training on the distribution station thermodynamic model to obtain a trained distribution station thermodynamic model.

[0042] Beneficial effects: The distribution room air-conditioning load regulation method based on model prediction provided by the present invention has the following advantages over the prior art: 1. The air-conditioning load of the distribution station is modeled based on an artificial intelligence method, and the air-conditioning cooling capacity is predicted, thereby realizing the air-conditioning load regulation of the distribution room, thereby improving the intelligence level of power grid regulation; 2. The thermodynamic model of the distribution station is trained from several important aspects, including spatial structure, temporal structure and power grid load demand, and the air-conditioning load regulation of the distribution station can be met without adding other parameters; 3. It can meet the situation of increasing distribution equipment as the intelligence level of the distribution network increases; 4. The influence of different factors on the air-conditioning load is preliminarily adjusted by feature fusion, and the influence of different factors on the air-conditioning load is further adjusted by adaptive adjustment fusion, so that the learning speed of the thermodynamic model of the distribution station is faster. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the implementation process of the present invention. DETAILED DESCRIPTION

[0044] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] like Figure 1 The figure shows a distribution room air conditioning load regulation method based on model prediction, which is executed by a distribution room air conditioning load regulation system. The distribution room air conditioning load regulation system includes a distribution station spatial structure feature extraction module, a data acquisition module, a distribution station thermodynamic model and a supervised training module.

[0046] The distribution station spatial structure feature extraction module first calculates the external wall orientation, external wall volume, internal volume, effective volume and window-to-wall ratio of the distribution station based on the geographical location and structural dimensions of the distribution station, and then extracts the distribution station spatial structure feature F based on the calculation results. space .

[0047] The data acquisition module collects the geographical location and structural dimensions of the distribution station and sends them to the distribution station spatial structure feature extraction module; collects quarterly information, weekly information, rest day information and daily scheduling time information, and sends them to the distribution station thermodynamic model; collects the surface temperature sequence of the distribution equipment, outdoor temperature, and sunshine information, and sends them to the distribution station thermodynamic model; collects the net load active power demand of the distribution network, and sends it to the distribution station thermodynamic model.

[0048] The distribution station thermodynamic model includes a distribution station temperature feature extraction module, a space fusion module, a time fusion module, an active demand fusion module, an adaptive adjustment fusion module and a detection head; the distribution station temperature feature extraction module is used to extract the distribution station temperature feature F tem ; The spatial fusion module is used to integrate the spatial structure characteristics F of the distribution station space and the distribution station temperature characteristic F tem Get spatial fusion features The time fusion module is used to extract the time feature F time , and integrate the temporal feature F time and the distribution station temperature characteristic F tem Get time fusion features The active demand fusion module is used to extract the active demand characteristics F of the net load of the distribution network. P , and integrate the active demand characteristics of the distribution network net load F P and the distribution station temperature characteristic F tem Get the active demand fusion feature F P ; The adaptive adjustment fusion module adjusts the spatial fusion features Temporal fusion features and active power demand fusion characteristics Perform multi-dimensional fusion to obtain multi-dimensional fusion features The detection head uses multi-dimensional fusion features Perform semantic segmentation to obtain the predicted value Q of air conditioning cooling capacity pre .

[0049] The supervised training module uses a loss function to perform supervised training on the distribution station thermodynamic model to obtain a trained distribution station thermodynamic model.

[0050] The present invention is further described below in conjunction with the specific implementation steps of the method of this case.

[0051] Step 1: Construct a module for extracting spatial structure features of distribution stations

[0052] Construct a distribution station spatial structure feature extraction module to extract the distribution station spatial structure feature F space ; The distribution station spatial structure characteristics F space It includes the exterior wall orientation, exterior wall volume, internal volume, effective volume and window-to-wall ratio of the distribution station calculated based on the geographical location and structural dimensions of the distribution station. The effective volume is the space after deducting the equivalent volume of all distribution equipment from the internal volume of the distribution station. The window-to-wall ratio is the ratio of the area of ​​the heat dissipation windows on each exterior wall of the distribution station to the area of ​​the exterior wall to which it belongs.

[0053] Step 2: Obtain historical data on heat loss and air conditioning load of power distribution equipment

[0054] Obtain historical data on heat loss and air conditioning load of power distribution equipment, including time characteristics F time , Distribution network net load active demand characteristics F P , distribution station temperature characteristics F tem and air conditioning cooling capacity label Q ture , the historical data is divided into a training data set and a test data set in proportion; the time feature F time Including quarterly information, weekly information, rest day information and daily scheduling time information, the distribution station temperature characteristic F tem Including distribution equipment surface temperature series, outdoor temperature, and sunshine information.

[0055] Step 3: Construct a thermodynamic model of a distribution station based on a neural network

[0056] The distribution station thermodynamic model is firstly fused with the spatial structure characteristics F of the distribution station through the spatial fusion module. space and the distribution station temperature characteristic F tem Perform feature fusion to obtain spatial fusion features By spatial fusion features Characterize the impact of the spatial structure of the distribution room on the air conditioning load; then use the time fusion module to fuzzy the time characteristics F time and the distribution station temperature characteristic F tem Perform feature fusion to obtain time fusion features Fusion of features through time Characterize the influence of time information on air conditioning load; then use the active demand fusion module to fusion the active demand characteristics of the net load of the distribution network F P and the distribution station temperature characteristic F tem Perform feature fusion to obtain active demand fusion features Fusion characteristics through active demand Characterize the impact of the net active demand of the distribution network on the air conditioning load; finally, the spatial fusion characteristics are adjusted by adaptively adjusting the fusion module Temporal fusion features and active power demand fusion characteristics Perform multi-dimensional fusion to obtain multi-dimensional fusion features Multi-dimensional fusion features Input to the detection head for semantic segmentation to obtain the predicted value Q of air conditioning cooling capacity pre .

[0057] The spatial fusion module, temporal fusion module and active demand fusion module in the thermodynamic model of the distribution station have the same structure, including a connection layer, a 1×1 convolution, a 3×3 convolution, a layer normalization layer and a Sigmoid activation function. The feature fusion process is expressed as:

[0058]

[0059]

[0060]

[0061] Among them: Cat represents the feature connection operation, represents a convolution with a 1×1 kernel, denotes a convolution with a 3×3 convolution kernel and Norm denotes a layer normalization operation.

[0062] Obtaining multi-dimensional fusion features through adaptive adjustment fusion module in the thermodynamic model of distribution station Here are the steps:

[0063] Step 31: First, spatial fusion features Temporal fusion features and active power demand fusion characteristics Add element by element to get Then F1 c Input to the channel attention unit and the spatial attention unit at the same time;

[0064] Step 32: In the channel attention unit, first focus on F1 along the channel dimension. c Perform a global average pooling operation Then, a 1×1 convolution is used to reduce the channel dimension, a ReLU activation function is used for nonlinear enhancement, and another 1×1 convolution is used to reduce the channel dimension. It is expressed as:

[0065]

[0066] Step 33: In the spatial attention unit, first focus F1 along the spatial dimension. c Perform a global average pooling operation And the global maximum pooling operation Then connect along the spatial dimension and input into a 7×7 convolution to get It is expressed as:

[0067]

[0068] in: represents a convolution with a 7×7 convolution kernel;

[0069] Step 34. and Add element by element to get Based on F1 c The corresponding channel pair Each channel of F1 is adjusted first along the channel dimension.c and Connect them, then rearrange each channel in an alternating manner through the channel shuffle operation, then pass the group convolution, and finally pass the Sigmoid activation function to obtain the fused weight map R, which is expressed as:

[0070]

[0071] Among them: CS represents the channel shuffle operation, represents a group convolution with a 7×7 convolution kernel;

[0072] Step 34, map the fusion weight R to the spatial fusion feature Temporal fusion features and active power demand fusion characteristics Performing element-wise multiplication, we get and

[0073] Step 34: Add each feature element by element, and then perform point-by-point convolution to obtain multi-dimensional fusion features. It is expressed as:

[0074]

[0075] in: Represents a point-wise convolution with a 1×1 kernel.

[0076] Step 4: Use loss function to supervise the training of the distribution station thermodynamic model

[0077] The mean absolute error function MAE and the mean absolute percentage error function MAPE are used as loss functions to supervise the distribution station thermodynamic model and obtain the trained distribution station thermodynamic model:

[0078]

[0079]

[0080] Where: K represents the number of samples in the training data set.

[0081] Step 5: Adjust the air conditioning load in the power distribution room based on the predicted value of air conditioning cooling capacity

[0082] The spatial structure characteristics of the distribution station and the test data set are input into the trained distribution station thermodynamic model to obtain the predicted value of air conditioning cooling capacity, and the air conditioning load of the distribution room is adjusted based on the predicted value of air conditioning cooling capacity.

[0083] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any form, and any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A method for regulating air conditioning load in a power distribution room based on model prediction, characterized in that: The steps include: Step 1: Construct a distribution station spatial structure feature extraction module to extract the distribution station spatial structure feature F space ; Step 2: Obtain historical data on heat loss and air conditioning load of power distribution equipment, including time characteristics F time , Distribution network net load active demand characteristics F P , distribution station temperature characteristics F tem and air conditioning cooling capacity label Q ture , divide the historical data into training data set and test data set in proportion; Step 3: Construct a thermodynamic model of the distribution station based on a neural network; first, use the spatial fusion module to quantify the spatial structure characteristics of the distribution station F. space and the distribution station temperature characteristic F tem Perform feature fusion to obtain spatial fusion features By spatial fusion features Characterize the influence of the spatial structure of the distribution room on the air conditioning load; then use the time fusion module to fuzzy the time characteristics F time and the distribution station temperature characteristic F tem Perform feature fusion to obtain time fusion features Fusion of features through time Characterize the influence of time information on air conditioning load; then use the active demand fusion module to fusion the active demand characteristics of the net load of the distribution network F P and the distribution station temperature characteristic F tem Perform feature fusion to obtain active demand fusion features Active demand fusion features Characterize the impact of the net active demand of the distribution network on the air conditioning load; finally, the spatial fusion characteristics are adjusted by adaptively adjusting the fusion module Temporal fusion features and active power demand fusion characteristics Perform multi-dimensional fusion to obtain multi-dimensional fusion features Multi-dimensional fusion features Input to the detection head for semantic segmentation to obtain the predicted value Q of air conditioning cooling capacity pre ; Step 4: Use the loss function to supervise the thermodynamic model of the distribution station and obtain a trained thermodynamic model of the distribution station; Step 5. Input the spatial structure characteristics of the distribution station and the test data set into the trained distribution station thermodynamic model to obtain the predicted value of air conditioning cooling capacity, and adjust the air conditioning load in the distribution room based on the predicted value of air conditioning cooling capacity.

2. The method for regulating the air conditioning load in a power distribution room based on model prediction according to claim 1 is characterized in that: In Step 1, the spatial structure feature F of the distribution station space It includes the exterior wall orientation, exterior wall volume, internal volume, effective volume and window-to-wall ratio of the distribution station calculated based on the geographical location and structural dimensions of the distribution station. The effective volume is the space after deducting the equivalent volume of all distribution equipment from the internal volume of the distribution station. The window-to-wall ratio is the ratio of the area of ​​the heat dissipation windows on each exterior wall of the distribution station to the area of ​​the exterior wall to which it belongs.

3. The method for regulating the air conditioning load in a power distribution room based on model prediction according to claim 1, characterized in that: In Step 2, the time feature F time Including quarterly information, weekly information, rest day information and daily scheduling time information, the distribution station temperature characteristic F tem Including distribution equipment surface temperature series, outdoor temperature, and sunshine information.

4. The method for regulating the air conditioning load in a power distribution room based on model prediction according to claim 1, characterized in that: In the Step 3, the structures of the spatial fusion module, the temporal fusion module and the active demand fusion module in the thermodynamic model of the distribution station are the same, including a connection layer, a 1×1 convolution, a 3×3 convolution, a layer normalization layer and a Sigmoid activation function. The feature fusion process is expressed as: Among them: Cat represents the feature connection operation, represents a convolution with a 1×1 kernel, denotes a convolution with a 3×3 convolution kernel and Norm denotes a layer normalization operation.

5. The method for regulating the air conditioning load in a power distribution room based on model prediction according to claim 1, characterized in that: In the Step 3, the multi-dimensional fusion features are obtained by adaptively adjusting the fusion module in the thermodynamic model of the distribution station. The steps include: Step 31: First, spatial fusion features Temporal fusion features and active power demand fusion characteristics Add element by element to get Then F1 c Input to the channel attention unit and the spatial attention unit at the same time; Step 32: In the channel attention unit, first focus on F1 along the channel dimension. c Perform a global average pooling operation Then, a 1×1 convolution is used to reduce the channel dimension, a ReLU activation function is used for nonlinear enhancement, and another 1×1 convolution is used to reduce the channel dimension. It is expressed as: Step 33: In the spatial attention unit, first focus F1 along the spatial dimension. c Perform a global average pooling operation And the global maximum pooling operation Then connect along the spatial dimension and input into a 7×7 convolution to get F3 c , expressed as: in: represents a convolution with a 7×7 convolution kernel; Step 34. and F3 c Add element by element to get Based on F1 c The corresponding channel pair Each channel of F1 is adjusted first along the channel dimension. c and Connect them, then rearrange each channel in an alternating manner through the channel shuffle operation, then pass the group convolution, and finally pass the Sigmoid activation function to obtain the fused weight map R, which is expressed as: Among them: CS represents the channel shuffle operation, represents a group convolution with a 7×7 convolution kernel; Step 34, map the fusion weight R to the spatial fusion feature Temporal fusion features and active power demand fusion characteristics Performing element-wise multiplication, we get and Step 34: Add each feature element by element, and then perform point-by-point convolution to obtain multi-dimensional fusion features. It is expressed as: in: Represents a point-wise convolution with a 1×1 kernel.

6. The method for regulating air conditioning load in a power distribution room based on model prediction according to claim 1, characterized in that: In the Step 4, the mean absolute error function MAE and the mean absolute percentage error function MAPE are used as loss functions to supervise the training of the distribution station thermodynamic model: Where: K represents the number of samples in the training data set.

7. The method for regulating air conditioning load in a power distribution room based on model prediction according to claim 1, characterized in that: The distribution room air conditioning load regulation method is executed by a distribution room air conditioning load regulation system, wherein the distribution room air conditioning load regulation system comprises a distribution station spatial structure feature extraction module, a data acquisition module, a distribution station thermodynamic model and a supervision training module; The distribution station spatial structure feature extraction module first calculates the external wall orientation, external wall volume, internal volume, effective volume and window-to-wall ratio of the distribution station based on the geographical location and structural dimensions of the distribution station, and then extracts the distribution station spatial structure feature F based on the calculation results. space ; The data acquisition module collects the geographical location and structural dimensions of the power distribution station and sends them to the power distribution station spatial structure feature extraction module; Collect quarterly information, weekly information, rest day information and daily scheduling time information, and send them to the thermodynamic model of the distribution station; collect the surface temperature sequence of the distribution equipment, outdoor temperature, and sunshine information, and send them to the thermodynamic model of the distribution station; collect the net load active power demand of the distribution network, and send them to the thermodynamic model of the distribution station; The distribution station thermodynamic model includes a distribution station temperature feature extraction module, a space fusion module, a time fusion module, an active demand fusion module, an adaptive adjustment fusion module and a detection head; The distribution station temperature feature extraction module is used to extract the distribution station temperature feature F tem ; The spatial fusion module is used to integrate the spatial structure characteristics F of the distribution station space and the distribution station temperature characteristic F tem Get spatial fusion features The time fusion module is used to extract the time feature F time , and integrate the temporal feature F time and the distribution station temperature characteristic F tem Get time fusion features The active demand fusion module is used to extract the active demand characteristics F of the net load of the distribution network. P , and integrate the active demand characteristics of the distribution network net load F P and the distribution station temperature characteristic F tem Get the active demand fusion feature F P ; The adaptive adjustment fusion module adjusts the spatial fusion features Temporal fusion features and active power demand fusion characteristics Perform multi-dimensional fusion to obtain multi-dimensional fusion features The detection head uses multi-dimensional fusion features Perform semantic segmentation to obtain the predicted value Q of air conditioning cooling capacity pre ; The supervised training module uses a loss function to perform supervised training on the distribution station thermodynamic model to obtain a trained distribution station thermodynamic model.

Images