A heating monitoring and analysis method based on big data

By using LSTM neural network and heat transfer model in the heating monitoring and analysis system, combined with autoregression algorithm, the problem of insufficient time and accuracy of heating load prediction in the prior art is solved, and heating with high spatio-temporal resolution is achieved Load forecasting improves energy utilization efficiency and system economy.

CN118296566BActive Publication Date: 2024-11-05YIDUO SMART ENERGY (DALIAN) CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202410414711.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-08
Publication Date
2024-11-05
Estimated Expiration
2044-04-08

AI Technical Summary

Technical Problem

The existing heating monitoring and analysis system has poor timeliness and accuracy in the prediction of heating loads in big data environments, and it is impossible to deeply explore the inherent relationship between heating loads and influencing factors, resulting in low energy utilization efficiency and economicality.

Method used

Using a heating monitoring and analysis method based on big data, the heating sub-region is divided by collecting house attribute data, and the LSTM neural network model is used to combine heat transfer model and autoregressive comprehensive moving average algorithm to establish a quantitative relationship model between heating load and influencing factors. , perform multi-step heating load prediction with high spatial and temporal resolution.

Benefits of technology

It significantly improves the accuracy and timeliness of heating load prediction, reduces energy waste, provides scientific basis for subsequent regulatory decisions, and improves the refinement and economicization of the heating system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118296566B_ABST
    Figure CN118296566B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of heating data processing, and discloses a heating monitoring and analysis method based on big data; collecting house attribute data, dividing a heating area into n heating sub-areas according to the house attribute data; collecting comprehensive heating data of the n heating sub-areas; inputting the comprehensive heating data into a pre-built LSTM neural network model to predict the heating load demand of the n heating sub-areas in the future t time period; making the heating system move towards a higher level of refinement, intelligence and economy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of heating data processing, and more specifically, to a heating monitoring and analysis method based on big data. Background Art

[0002] The patent with authorization announcement number CN115964668A discloses a heating monitoring and analysis method, device, equipment and medium based on big data; the method includes: obtaining historical heating information and heating equipment information, performing data analysis on the historical heating information, obtaining abnormal equipment information, determining the first equipment node information and the first equipment associated information that need to be managed, managing and tracing the first equipment node information, obtaining the second equipment node information and the second equipment associated information, and then constructing the first equipment node group information, associating and constructing the first equipment node group information, obtaining the second equipment node group information, accessing each data node information in the second equipment node group information, obtaining equipment detail information, comparing the equipment detail information with the abnormal equipment information, and obtaining abnormal cause information, which has the effect of improving the efficiency of solving heating abnormalities.

[0003] However, the current heating monitoring and analysis system uses a simple empirical model to predict heating load, which is difficult to adapt to changes in demand under the big data environment; the timeliness and accuracy of the prediction results are poor, and fail to provide a scientific basis for subsequent regulation; the empirical model established by the traditional method cannot deeply explore the internal connection and quantitative relationship between the heating load and the actual influencing factors, and the prediction deviation is large. Different treatment and modeling for different regions and building types are also insufficient; the existing system lacks scientific guidance and optimization for energy use, and directly sets the operating status and power output of heating equipment based on experience, which often results in a state of excess equipment capacity; the energy utilization efficiency and economy of the entire heating system are low.

[0004] In view of this, the present invention proposes a heating monitoring and analysis method based on big data to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: A heating monitoring and analysis method based on big data, comprising: S1, collecting house attribute data, and dividing the heating area into n heating sub-areas according to the house attribute data;

[0006] S2, collecting comprehensive heating data of n heating sub-areas;

[0007] S3. Input the comprehensive heating data into the pre-built LSTM neural network model to analyze the heating load demand of n heating sub-areas in the future t time period.

[0008] Furthermore, the house attribute data includes house area, house type, house wall insulation rate, house orientation and house floor height;

[0009] The method for obtaining the thermal insulation rate of the house wall includes:

[0010] Use an infrared thermal imager to scan the wall of the house to obtain a thermal map of the wall surface temperature; during scanning, ensure that the distance between the infrared thermal imager and the wall is within B meters; while collecting the thermal map of the wall temperature, use a temperature sensor to collect the ambient temperature within a range of L meters around the house wall; through an image processing algorithm, define an area of ​​interest on the thermal map of the wall surface temperature and extract the average temperature within the area of ​​interest; select an area of ​​interest for every 1 square meter of the house wall scanned; repeat the scan C times to obtain sample data of the wall surface temperature;

[0011] Establish a heat transfer model for the wall based on the material data of the wall material; combine the heat transfer model with the sample data of the wall surface temperature and the ambient temperature, and use the optimization algorithm to inversely solve the wall insulation rate;

[0012] The heat transfer model is established in the following manner:

[0013] Check the product manual or test report of the wall material to determine the material data of the wall material; the material data includes density, specific heat capacity and thermal conductivity of the material;

[0014] A one-dimensional steady-state heat conduction model is established. Using the one-dimensional control volume method, the wall is equivalent to E two-sided control units, each of which contains four boundary surfaces. The energy conservation equation of the two-sided control unit is defined as:

[0015] Among them, ΔW is the thickness of the two-sided control unit; ω is the thermal conductivity of the material, D is the boundary surface area, Q W and Q' W are the heat flux densities at the left and right interfaces respectively;

[0016] Discretize the energy conservation equation to get the heat transfer model T1=U1·T a +R1;

[0017] Among them, T1, T a are the wall surface temperature sample data and the ambient temperature respectively; U1 is the coefficient matrix, and R1 is the heat flow vector;

[0018] The method of combining the heat transfer model with the wall surface temperature sample data and the ambient temperature and using the optimization algorithm to inversely solve the wall insulation rate includes:

[0019] Substitute the ambient temperature into the wall heat transfer model and initialize the value range of the wall insulation rate [O min ,Omax ];

[0020] In [O min ,O max ] select Y insulation rates; construct a corresponding coefficient matrix according to each insulation rate, substitute the coefficient matrix into the heat transfer model, and calculate the wall surface temperature;

[0021] Define the mean square error between the temperature output by the heat transfer model and the wall surface temperature sample data;

[0022] Mean square error in, is the temperature output by the heat transfer model; T Hi is the Hi-th wall surface temperature sample data, Hj is the total number of wall surface temperature sample data;

[0023] Calculate the mean square error between the model output temperature and the wall surface temperature sample data; use the genetic algorithm to search for the insulation rate to minimize the mean square error and obtain the final insulation rate value.

[0024] Furthermore, the method of dividing the heating area into n heating sub-areas includes:

[0025] The autoregressive integrated moving average algorithm is used to establish a quantitative relationship model between the attribute and the load; the quantitative relationship model is established in the following manner:

[0026] Extract the optimal relationship data from the house attribute data, use the optimal relationship data as the exogenous variable of the autoregressive integrated moving average algorithm, and use the heating load corresponding to the optimal relationship data as the dependent variable of the autoregressive integrated moving average algorithm;

[0027] Initialize the hyperparameters of the autoregressive integrated moving average algorithm, bring the exogenous variables and dependent variables into the autoregressive integrated moving average algorithm, and use the maximum likelihood estimation method for training; obtain the constant term, AR coefficient, and MA coefficient of the autoregressive integrated moving average algorithm; and obtain the quantitative relationship model by analyzing the coefficients of each exogenous variable;

[0028] The house is initially divided into v cubes, and the average value of the heating load in each cube is calculated according to the quantitative relationship model as the average value variable; the world coordinates of the center point of each cube are recorded; the Euclidean distance between the cubes is calculated according to the world coordinates of the center point of each cube;

[0029] Pearson correlation coefficients were calculated based on the Euclidean distances between cubes and the mean variables;

[0030] Take all the divided v cubes as sample areas, calculate the Pearson correlation coefficients between the v cubes, and form a v×v symmetric matrix R; the elements in the kth row and hth column of the symmetric matrix R represent the Pearson correlation coefficients between cube k and cube h;

[0031] For each sample region u, calculate the average Pearson correlation coefficient a(u) between it and other samples in the same cluster, and the average Pearson correlation coefficient b(u) between it and all samples in the nearest cluster;

[0032] Calculate the silhouette coefficient of the sample area

[0033] Calculate the overall silhouette coefficient S = mean(s(u)), gradually reduce K, and repeatedly calculate S; record the K that makes S take the maximum value as the optimal number of categories; the corresponding clustering result is the best area division; merge the sample areas divided into the same cluster at this time to obtain the heating sub-area; K at this time is the number of heating sub-areas n.

[0034] Furthermore, the optimal relationship data is extracted by obtaining preliminary model variables through correlation analysis; performing lasso regression analysis on the preliminary model variables to obtain the optimal relationship data;

[0035] Ways to obtain preliminary model variables through correlation analysis include:

[0036] Initialize a correlation matrix; calculate the correlation coefficients between the elements in the correlation matrix; the elements in the correlation matrix are the property data of the house;

[0037] Use color shades to represent r ij The correlation heat map between variables is drawn according to the numerical value of ; the correlation heat map consists of m grids; each grid is marked with the value of the correlation coefficient;

[0038] Identify the darker grids in the correlation heat map, find the variable pairs whose correlation coefficients are higher than the preset threshold, that is, there is multicollinearity between the variable pairs; remove the variables with multicollinearity; after removing the variables, recalculate the correlation coefficient matrix; draw a new heat map to determine whether there are still dark grids; if so, identify and remove the variables with multicollinearity again; until there are no multicollinear variables in the correlation heat map; the variables retained at this time are the preliminary model variables;

[0039] The calculation formula of the correlation coefficient is: Among them, r ij is the element in the i-th row and j-th column of the correlation matrix, representing the variable X i and X jThe correlation coefficient between Cov(X i ,X j ) is the variable X i and X j The covariance between i For variable X i The standard deviation of j For variable X j The standard deviation of .

[0040] Furthermore, the method of performing lasso regression analysis on the preliminary model variables to obtain the optimal relationship data includes:

[0041] A penalty term is added to the loss function of the preset linear regression model to obtain a penalty loss function; the penalty loss function is used to perform penalty processing to obtain the next-step model variable;

[0042] The formula of the linear regression model is: Y=Zw+e, where Y is the characteristic variable obtained by the target; Z is the input characteristic matrix; the characteristic matrix is ​​the matrix composed of the preliminary model variables; w is the regression coefficient, and e is the random error;

[0043] The penalty loss function is: Where α is a hyperparameter of the regularization strength; I is the sum index of the regression coefficient w in the L1 norm; p is the total number of preliminary model variables; |w I | represents the L1 norm of the I-th regression coefficient w;

[0044] The punishment methods include:

[0045] Set an initial α to obtain a sparse regression coefficient vector; treat variables whose regression coefficients are penalized to 0 as non-critical variables; retain variables whose regression coefficients are not 0; this is the optimal relationship data.

[0046] Furthermore, the comprehensive heating data includes heating data and meteorological parameter data;

[0047] Heating data include pipe network flow, supply and return water temperature, unit start and stop status and fuel consumption; meteorological parameter data include outdoor temperature, humidity, wind speed and direction;

[0048] The start and stop status of the unit and the wind direction are numerically encoded; the normal data interval corresponding to the heating data is defined, and the abnormal values ​​in the historical heating data, that is, the data beyond the normal data interval, are removed according to the normal data interval; and the data of different dimensions are mapped to the same numerical range.

[0049] Furthermore, the LSTM neural network model is constructed in the following manner:

[0050] Initialize the basic architecture of the LSTM neural network model, which includes an input layer, two LSTM units, a Dropout layer, a fully connected layer, a learnable parameter activation function, a comprehensive loss function, an optimizer, a regularizer, and an output layer; the output layer includes n nodes;

[0051] Each LSTM unit consists of a memory unit and three control gates; the three control gates are input gate, forget gate and output gate; a fully connected layer is added between LSTM units;

[0052] The Dropout layer randomly discards a certain proportion of nodes from the LSTM neural network model;

[0053] The regularizer uses L2 norm regularization to control the LSTM neural network model to prevent overfitting; the output layer including n nodes is used to output the heating load demand in the future t time period corresponding to n heating sub-areas;

[0054] Collect comprehensive heating data of n heating sub-areas over a period of time in history, and collect the heating load corresponding to the comprehensive heating data; construct the data into training data; divide the training data into a training set and a validation set;

[0055] The training set is output to the LSTM neural network model, the heating comprehensive data is received as the input layer, the heating load is used as the output of the output layer, and the n heating sub-areas correspond to the n nodes of the output layer; during the training process, when the function value of the comprehensive loss function does not decrease during the process of P consecutive iterations, the training is completed, and the trained LSTM neural network model is obtained;

[0056] The collected comprehensive heating data of n heating sub-areas are input into the trained LSTM neural network model to obtain the heating load demand of the n heating sub-areas in the future t time period.

[0057] Furthermore, the learnable parameter activation function Where f is the input data; p1 is the learnable slope parameter; p2 is the learnable offset parameter;

[0058] The comprehensive loss function Among them, q is the quantile, which takes values ​​between (0, 1); y true is the real target value; is the lower quantile of the predicted distribution; is the upper quantile of the predicted distribution; | | + is a positive function.

[0059] Furthermore, the optimizer adopts an adaptive optimizer that adaptively adjusts the learning rate;

[0060] The optimization formula of the adaptive optimizer is: in, is the parameter of the lth layer of the LSTM neural network model when the number of training iterations is λ+1; is the parameter of the lth layer of the LSTM neural network model when the number of training iterations is λ; η l is the learning rate of the lth layer of the LSTM neural network model; v λ is the exponential moving average of the squared gradient for training iterations λ; is a constant; m λ,l is the parameter gradient for the training iteration number λ;

[0061] like but in is the value of the comprehensive loss function with the number of training iterations λ; is the value of the comprehensive loss function when the number of training iterations is λ-1; A is the reduction parameter.

[0062] A heating monitoring and analysis system based on big data, which is used to implement the heating monitoring and analysis method based on big data, comprises: a collection and division module, which is used to collect house attribute data, and divide the heating area into n heating sub-areas according to the house attribute data;

[0063] The heating collection module is used to collect comprehensive heating data of n heating sub-areas;

[0064] The load demand prediction module is used to input the comprehensive heating data into the pre-built LSTM neural network model to predict the heating load demand of n heating sub-areas in the future t time period. The modules are connected by wired and / or wireless means to realize data transmission between modules.

[0065] Technical effects and advantages of a heating monitoring and analysis method based on big data of the present invention:

[0066] The present invention greatly improves the accuracy and timeliness of heating load prediction. By establishing a precise quantitative relationship between heating load and influencing factors and introducing a deep learning model in a big data environment, it realizes multi-step heating load prediction with high temporal and spatial resolution for the heating area, providing a reliable basis for subsequent regulation and control decisions; significantly reduces the energy waste of the heating system; through the heating load prediction results in multiple regions and multiple time periods, it avoids the waste of resources caused by excessive heating and reduces the system operating cost; through a complete data preprocessing process, effectively filters abnormal data, unifies dimensions, etc., it enhances the adaptability of the prediction model; at the same time, the proposed deep learning model training scheme also enhances the generalization ability of the model; realizes accurate digital modeling of complex heating scenarios, so that the heating system moves towards a higher level of refinement, intelligence, and economy. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a schematic diagram of a heating monitoring and analysis method based on big data of the present invention;

[0068] Figure 2 This is a schematic diagram of a 3x3 matrix in Embodiment 1 of the present invention;

[0069] Figure 3 A schematic diagram of a heating monitoring and analysis system based on big data of the present invention;

[0070] Figure 4 It is a schematic diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] Example 1

[0073] See also Figure 1 As shown, the heat supply monitoring and analysis method based on big data described in this embodiment includes:

[0074] S1. Collecting house attribute data, and dividing the heating area into n heating sub-areas according to the house attribute data;

[0075] S2, collecting comprehensive heating data of n heating sub-areas;

[0076] S3. Input the comprehensive heating data into the pre-built LSTM neural network model to analyze the heating load demand of n heating sub-areas in the future t time period.

[0077] Furthermore, the house attribute data includes house area, house type, house wall insulation rate, house orientation and house floor height;

[0078] The house area is extracted from the CAD drawing of the house; at the same time, the house type and the house floor height are determined according to the CAD drawing, and the house type is numerically numbered, for example, a family residence is numbered 1, a commercial office space is numbered 2, a public indoor space is numbered 3, and so on;

[0079] The method for obtaining the thermal insulation rate of the house wall includes:

[0080] Use an infrared thermal imager to scan the walls of the house and obtain a thermal map of the wall surface temperature. When scanning, ensure that the distance between the infrared thermal imager and the wall is within B meters. While collecting the thermal map of the wall temperature, use a temperature sensor to collect the ambient temperature within L meters around the house wall. The temperature sensor is placed in a shaded and ventilated place to prevent the heat generated by the device itself from affecting the measurement.

[0081] Through the image processing algorithm, the region of interest is delineated on the wall surface temperature heat map, and the average temperature in the region of interest is extracted; one region of interest is selected for each scan of 1 square meter of the house wall; the scan is repeated C times to obtain the wall surface temperature sample data;

[0082] Establish a heat transfer model for the wall based on the material data of the wall material; combine the heat transfer model with the sample data of the wall surface temperature and the ambient temperature, and use the optimization algorithm to inversely solve the wall insulation rate;

[0083] It should be noted that the method of defining the region of interest on the wall surface temperature heat map and extracting the average temperature in the region of interest includes:

[0084] Use OpenCV or other computer vision algorithm libraries to read thermal images and perform denoising, including median filtering and bilateral filtering algorithms to remove image noise; set the temperature range, for example, focus on the 20℃~35℃ area, perform image binarization processing, and highlight the temperature area of ​​interest; use Canny edge detection or Sobel edge detection algorithm to detect the edge of the area of ​​interest; through contour traversal, use rectangular contour fitting to select the minimum circumscribed rectangle of the area of ​​interest as the area boundary; calculate the temperature values ​​of all pixels within the area boundary;

[0085] The arithmetic mean of the temperature values ​​of all pixels in the rectangular area is calculated as the average temperature of the area of ​​interest; the average temperature of each area of ​​interest is repeatedly extracted every time 1 square meter of wall is scanned.

[0086] The house orientation is the orientation of the side with more windows in the house. The house orientation is numerically numbered; the house orientation of the east is numbered 1; the house orientation of the west is numbered 2; the house orientation of the south is numbered 3; the house orientation of the north is numbered 4;

[0087] The heat transfer model is established in the following manner:

[0088] Check the product manual or test report of the wall material to determine the material data of the wall material; the material data includes density, specific heat capacity and thermal conductivity of the material;

[0089] A one-dimensional steady-state heat conduction model is established. Using the one-dimensional control volume method, the wall is equivalent to E two-sided control units, each of which contains four boundary surfaces. The energy conservation equation of the two-sided control unit is defined as:

[0090] Among them, ΔW is the thickness of the two-sided control unit; ω is the thermal conductivity of the material, D is the boundary surface area, Q W and Q' W are the heat flux densities at the left and right interfaces respectively;

[0091] The heat flux density of the left interface and the right interface is the quantity introduced when defining the energy conservation equation of each two-surface control unit;

[0092] It is obtained by the following mathematical method:

[0093] ρ.cp.Vp.tx=q left ”D+q right ”D; where ρ is density; cp is specific heat capacity; Vp is the volume of the two-sided control unit; tx is the rate of change of temperature over time; q left and q right represent the heat flux density on the left and right interfaces respectively;

[0094] Discretize the energy conservation equation to get the heat transfer model T1=U1·T a +R1;

[0095] Among them, T1, T a are the wall surface temperature sample data and the ambient temperature respectively; U1 is the coefficient matrix, and R1 is the heat flow vector; the coefficient matrix can be obtained through the discrete energy conservation equation;

[0096] The method of combining the heat transfer model with the wall surface temperature sample data and the ambient temperature and using the optimization algorithm to inversely solve the wall insulation rate includes:

[0097] Substitute the ambient temperature into the wall heat transfer model and initialize the value range of the wall insulation rate [O min ,O max ];

[0098] In [O min ,O max ] select Y insulation rates; construct a corresponding coefficient matrix according to each insulation rate, substitute the coefficient matrix into the heat transfer model, and calculate the wall surface temperature;

[0099] Calculate the mean square error between the output temperature of the heat transfer model and the measured temperature;

[0100] Mean square error in, is the model output temperature; T Hi is the Hi-th wall surface temperature sample data, Hj is the total number of wall surface temperature sample data;

[0101] Calculate the mean square error between the model output temperature and the wall surface temperature sample data; use the genetic algorithm to search for the insulation rate to minimize the mean square error and obtain the final insulation rate value.

[0102] Furthermore, the method of dividing the heating area into n heating sub-areas includes:

[0103] The autoregressive integrated moving average algorithm is used to analyze the impact of house attribute data on historical heating load and establish a quantitative relationship model between attributes and load.

[0104] The method of establishing the quantitative relationship model includes:

[0105] Extract the optimal relationship data from the house attribute data, use the optimal relationship data as the exogenous variable of the autoregressive integrated moving average algorithm, and use the heating load corresponding to the optimal relationship data as the dependent variable of the autoregressive integrated moving average algorithm;

[0106] The method for obtaining the heating load includes:

[0107] Install supply and return water thermometers and flow meters at the heating interface, and calculate the instantaneous heating load according to the relationship between mass flow and water temperature difference;

[0108] The calculation formula for instantaneous heating load is: QW = mm.cc.(Ty supply -Ty return ); where QW is the heating load; mm is the mass flow rate of water; cc is the specific heat capacity of water; Ty supply is the water inlet temperature; Ty is the water outlet temperature;

[0109] Initialize the hyperparameters of the autoregressive integrated moving average algorithm, bring the exogenous variables and dependent variables into the autoregressive integrated moving average algorithm, and use the maximum likelihood estimation method for training; obtain the constant term, AR coefficient, and MA coefficient of the autoregressive integrated moving average algorithm; and obtain the quantitative relationship model by analyzing the coefficients of each exogenous variable;

[0110] In order to better explain the above process, for example, when the autoregressive integrated moving average algorithm (ARIMA (p, q, d) model) is established and exogenous variables 1 and 2 are introduced, the algorithm is expressed as:

[0111] y t =c+a1·x 1,t +b1·x 2,t +φ1·y t-1 +φ2·y t-2+ε t +θ1·ε t-1 +θ2·ε t-2 ;

[0112] In the formula, y t is the heating load at time point t; x 1,t is the exogenous variable 1 at time point t; x 2,t is the exogenous variable 2 at time point t; c is the constant term; φ1 and φ2 are AR coefficients; θ1 and θ2 are MA coefficients; a1 is the coefficient of exogenous variable 1; b1 is the coefficient of exogenous variable 2; ε t is the error term at time point t. The error term represents the random part that cannot be explained by the algorithm, that is, the influencing factors or random fluctuations that the model fails to capture; y t-1 is the heating load at time point t-1; y t-2 is the heating load at time point t-2; ε t-1 is the error term at time point t-1; ε t-2 is the error term at time point t-2;

[0113] For example, we obtain a1=0.5, b1=0.2; perform hypothesis test on the coefficients, specifically, assume that the sample H0 is a1=0; calculate the p-value in the autoregressive integrated moving average algorithm, if the p-value is less than 0.05, reject the sample H0, indicating that the coefficient of the exogenous variable 1 is significant;

[0114] The direction of the impact is determined by the positive or negative coefficient. If the coefficient of exogenous variable 1 is positive, then exogenous variable 1 is positively correlated with the heating load. The elasticity coefficient is calculated to determine the magnitude of the impact. If the elasticity of exogenous variable 1 is 0.5, it means that if exogenous variable 1 increases by 1%, the heating load increases by 0.5%.

[0115] The coefficient results show that both exogenous variables 1 and 2 are positive factors affecting heating load, among which the influence of exogenous variable 1 is slightly greater than that of exogenous variable 2;

[0116] The quantitative relationship model is: y=c+0.5·x1+0.2·x2; where y is the heating load; x1 is exogenous variable 1, and x2 is exogenous variable 2.

[0117] The optimal relationship data is extracted by obtaining preliminary model variables through correlation analysis; performing lasso regression analysis on the preliminary model variables to obtain the optimal relationship data;

[0118] Ways to obtain preliminary model variables through correlation analysis include:

[0119] Initialize a correlation matrix; calculate the correlation coefficients between the elements in the correlation matrix; the elements in the correlation matrix are the property data of the house; for example, there are three variables of the property data of the house, namely X1 is the house area; X2 is the house floor height; X3 is the house wall insulation rate; calculate the correlation coefficients between these three variables and put them into a 3x3 matrix, such as Figure 2 As shown in the 3x3 matrix; the 1 on the diagonal of this matrix represents that each variable is 100% correlated with itself; each other element r12 represents the correlation coefficient between X1 and X2, r13 represents the correlation coefficient between X1 and X3, and so on; the range of the correlation coefficient is between -1 and 1, a positive value represents a positive correlation, a negative value represents a negative correlation, and 0 represents no correlation; the larger the absolute value, the stronger the correlation; for example, if r12 = 0.9, it means that X1 and X2 are strongly positively correlated;

[0120] The calculation formula of the correlation coefficient is: Among them, r ij is the element in the i-th row and j-th column of the correlation matrix, representing the variable X i and X j The correlation coefficient between Cov(X i ,X j ) is the variable X i and X j The covariance between i For variable X i The standard deviation of j For variable X j The standard deviation of

[0121] Use color shades to represent r ij The correlation heat map between variables is drawn according to the numerical value of ; the correlation heat map is composed of m grids; each grid is marked with the value of the correlation coefficient for easy identification; the size, title and axis label elements of the correlation heat map are adjusted to ensure that the heat map is highly readable;

[0122] Identify the darker grids in the correlation heat map, find the variable pairs whose correlation coefficients are higher than the preset threshold, that is, there is multicollinearity between the variable pairs; remove the variables with multicollinearity; after removing the variables, recalculate the correlation coefficient matrix; draw a new heat map to determine whether there are still dark grids; if so, identify and remove the variables with multicollinearity again; until there are no multicollinear variables in the correlation heat map; at this time, there is basically no multicollinearity between the retained variables;

[0123] It needs to be explained that multicollinearity means that two variables represent similar information and change trends in the sample space; if they enter the model at the same time, the regression coefficient will be unstable, affecting the explanatory power and predictive power of the model;

[0124] The threshold value may be preset in the following manner:

[0125] Collect historical data and calculate the correlation coefficient matrix R between variables; set the preset threshold interval, such as [0.7, 0.9]; select several candidate thresholds within this interval, such as 0.75, 0.8, 0.85; for each candidate threshold; find the variable pairs whose correlation coefficients are greater than this candidate threshold, and these variables have multicollinearity; remove these variables to obtain variable subsets; use the variable subsets to build models and compare model effects; compare the effects of the models under different candidate thresholds, and select the parameters corresponding to the best model as the preset threshold; use the selected preset threshold to find multicollinear variables and remove the variables; repeat the above process until there are no multicollinear variables.

[0126] The method of performing lasso regression analysis on the preliminary model variables to obtain the optimal relationship data includes:

[0127] A penalty term is added to the loss function of the preset linear regression model to obtain a penalty loss function; the penalty loss function is used to perform penalty processing to obtain the next-step model variable;

[0128] The formula of the linear regression model is: Y=Zw+e, where Y is the characteristic variable obtained by the target; Z is the input characteristic matrix; the characteristic matrix is ​​the matrix composed of the preliminary model variables; w is the regression coefficient, and e is the random error; the regression coefficient represents the weight or contribution of each input feature when predicting the target variable (characteristic variable);

[0129] The penalty loss function is: Where α is a hyperparameter of the regularization strength; I is the sum index of the regression coefficient w in the L1 norm; p is the total number of preliminary model variables; |w I | represents the L1 norm of the I-th regression coefficient w;

[0130] The punishment methods include:

[0131] Set an initial α to obtain a sparse regression coefficient vector; treat variables whose regression coefficients are penalized to 0 as non-critical variables; retain those variables whose regression coefficients are not 0; this is the optimal relationship data.

[0132] The house is initially divided into v cubes, and the average value of the heating load in each cube is calculated according to the quantitative relationship model as the average value variable; the world coordinates of the center point of each cube are recorded; the Euclidean distance between the cubes is calculated according to the world coordinates of the center point of each cube;

[0133] Pearson correlation coefficients were calculated based on the Euclidean distances between cubes and the mean variables;

[0134] The formula for calculating the Pearson correlation coefficient is: Among them, R kh is the Pearson correlation coefficient between cube k and cube h; l k is the average value variable of cube k; l h is the average value variable of the cube h; μ k is the average value of the heating load data of all cubes around cube k except cube k; μ h is the average value of the heating load data of all cubes around cube h except h; σ k is the standard deviation of the heating load data of all cubes around cube k except cube k; σ h is the standard deviation of the heating load data of all cubes around cube h except h;

[0135] Take all the divided v cubes as sample areas, calculate the Pearson correlation coefficients between the v cubes, and form a v×v symmetric matrix R; the elements in the kth row and hth column of the symmetric matrix R represent the Pearson correlation coefficients between cube k and cube h;

[0136] Initialize the parameters of the K-means algorithm, set the number of clustering categories K, and randomly select K regions as the initial clustering centers;

[0137] Traverse all sample areas, calculate the Pearson correlation coefficient between each sample area and K cluster centers; classify each sample area into the cluster with the largest Pearson correlation coefficient; and recalculate the center of each cluster, that is, the mean of the Pearson correlation coefficients between each sample area and the cluster center; repeat until the cluster center no longer changes and the K-means algorithm converges;

[0138] For each sample region u, calculate the average Pearson correlation coefficient a(u) between it and other samples in the same cluster, and the average Pearson correlation coefficient b(u) between it and all samples in the nearest cluster;

[0139] Calculate the silhouette coefficient of the sample area The value range of the silhouette coefficient is [-1, 1]. The larger the value, the greater the correlation between the sample area and the cluster it belongs to, and the better the clustering effect.

[0140] Calculate the overall silhouette coefficient S = mean(s(u)), which is the mean of the silhouette coefficients of all sample areas; its value range is [-1, 1], and the larger the value, the better the clustering effect; gradually reduce K and repeatedly calculate S; usually as K decreases, S first increases and then decreases.

[0141] Record K that makes S take the maximum value as the optimal number of categories; the corresponding clustering result is the best area division; merge the sample areas divided into the same cluster at this time to obtain the heating sub-area; K at this time is the number of heating sub-areas n.

[0142] Furthermore, the comprehensive heating data includes heating data and meteorological parameter data;

[0143] Heating data includes pipe network flow, supply and return water temperature, unit start and stop status and fuel consumption; meteorological parameter data includes outdoor temperature, humidity, wind speed and wind direction; all of the above data can be obtained through sensors or direct reading;

[0144] Numerical encoding of the start / stop status of the unit and the wind direction, for example, one-hot encoding; defining a corresponding normal data interval for historical heating data, and removing abnormal values ​​in the historical heating data according to the normal data interval, that is, data that exceeds the normal data interval; mapping data of different dimensions to the same numerical range;

[0145] Furthermore, the LSTM neural network model is constructed in the following manner:

[0146] Initialize the basic architecture of the LSTM neural network model, which includes an input layer, two LSTM units, a Dropout layer, a fully connected layer, a learnable parameter activation function, a comprehensive loss function, an optimizer, a regularizer, and an output layer; the output layer includes n nodes;

[0147] The input layer organizes the meteorological parameter data into a format that can be processed by the LSTM neural network model and contains 4 nodes; for example, if the meteorological parameter data of each time step is a fixed-dimensional vector, then the input layer output is a three-dimensional tensor of (batch_size, timesteps, input_dim); batch_size represents the number of training samples in each batch, timesteps represents the number of time steps, and input_dim represents the dimension of the input data at each time step;

[0148] Each LSTM unit consists of a memory unit and three control gates; the three control gates are input gate, forget gate and output gate; a fully connected layer is added between LSTM units;

[0149] The Dropout layer randomly discards a certain proportion of nodes from the LSTM neural network model temporarily, which can effectively prevent overfitting;

[0150] The learnable parameter activation function Among them, f is the input data; p1 is a learnable slope parameter that controls the slope of the linear transformation of f. The value of p1 affects the gradient of the Sigmoid function in the input space, thereby affecting the response speed of the Sigmoid function to the input; learning p1 can make the model adapt to different ranges of activation values ​​in the data; p2 is a learnable offset parameter that controls the offset of the Sigmoid function, so that the model can adapt to different baselines or offsets in the input data; learning p2 can make the model better model the overall pattern of the data;

[0151] The specific values ​​of p1 and p2 are obtained through model training; in deep learning, training is performed through the back-propagation algorithm and optimizer, which adjusts the parameters in the network so that the model achieves better performance on the training data;

[0152] Uncertainty modeling is introduced into the loss function to obtain the comprehensive loss function Loss q ;

[0153] Among them, q is the quantile, which indicates the quantile level in the loss calculation; q takes a value between (0, 1), which determines whether the loss is calculated for the lower quantile or the upper quantile; y true is the real target value, indicating the actual heating load; is the lower quantile of the predicted distribution; is the upper quantile of the predicted distribution; | | + is the positive function; it is used to ensure that only when the difference between the predicted value and the true value is positive will it contribute to the loss;

[0154] The optimizer adopts an adaptive optimizer that adaptively adjusts the learning rate;

[0155] The optimization formula of the adaptive optimizer is: in, is the parameter of the lth layer of the LSTM neural network model when the number of training iterations is λ+1; is the parameter of the lth layer of the LSTM neural network model when the number of training iterations is λ; η l is the learning rate of the lth layer of the LSTM neural network model, and different learning rates are selected according to different layers; v λ is the exponential moving average of the squared gradient for training iterations λ; is a small constant that prevents the denominator from being zero; m λ,l is the parameter gradient for the training iteration number λ;

[0156] like but in is the value of the comprehensive loss function with the number of training iterations λ; is the value of the comprehensive loss function when the number of training iterations is λ-1; A is the reduction parameter;

[0157] The regularizer uses L2 norm regularization to control the complexity of the LSTM neural network model and prevent overfitting;

[0158] The n output layers are used to output the heating load demand in the future t time period corresponding to the n heating sub-areas;

[0159] Collect the comprehensive heating data of n heating sub-areas for a period of time in history, and collect the heating load corresponding to the comprehensive heating data; and construct the training data; divide the training data into a training set and a validation set; the specific division ratio can be determined according to the actual situation, for example, 80% training set and 20% validation set;

[0160] The training set is output to the LSTM neural network model, the heating comprehensive data is received as the input layer, the heating load is used as the output of the output layer, and the n heating sub-areas correspond to the n nodes of the output layer; during the training process, when the function value of the comprehensive loss function does not decrease during the process of P consecutive iterations, the training is completed, and the trained LSTM neural network model is obtained;

[0161] The collected comprehensive heating data of n heating sub-areas are input into the trained LSTM neural network model, and the heating load demand of the n heating sub-areas in the future t time period is analyzed.

[0162] This embodiment greatly improves the accuracy and timeliness of heating load prediction. By establishing a precise quantitative relationship between heating load and influencing factors and introducing a deep learning model in a big data environment, it realizes multi-step heating load prediction with high temporal and spatial resolution for the heating area, providing a reliable basis for subsequent control decisions; significantly reduces the energy waste of the heating system; through the heating load prediction results in multiple regions and multiple time periods, it avoids the waste of resources caused by excessive heating and reduces the system operation cost; through a complete data preprocessing process, it effectively filters abnormal data and unifies dimensions, etc., thereby enhancing the adaptability of the prediction model; at the same time, the proposed deep learning model training scheme also strengthens the generalization ability of the model; it realizes accurate digital modeling of complex heating scenarios, so that the heating system moves towards a higher level of refinement, intelligence, and economy.

[0163] Example 2

[0164] See also Figure 2 As shown, the part not described in detail in this embodiment is described in the description of embodiment 1, and a heating monitoring and analysis system based on big data is provided, including: a collection and division module, which is used to collect house attribute data, and divide the heating area into n heating sub-areas according to the house attribute data;

[0165] The heating collection module is used to collect comprehensive heating data of n heating sub-areas;

[0166] The load demand prediction module is used to input the comprehensive heating data into the pre-built LSTM neural network model to analyze the heating load demand of n heating sub-areas in the future t time period; each module is connected by wired and / or wireless means to realize data transmission between modules.

[0167] Example 3

[0168] See also Figure 3 As shown, this embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the operation mode of the above-mentioned heating monitoring and analysis method based on big data is implemented.

[0169] Since the electronic device introduced in this embodiment is an electronic device used to implement a method for monitoring and analyzing heating supply based on big data in the embodiment of this application, based on the method for monitoring and analyzing heating supply based on big data introduced in the embodiment of this application, a person skilled in the art can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as a person skilled in the art implements an electronic device used in a method for monitoring and analyzing heating supply based on big data in the embodiment of this application, it belongs to the scope of protection of this application.

[0170] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0171] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A heating monitoring and analysis method based on big data, characterized in that: include: S1. Collecting house attribute data, and dividing the heating area into n heating sub-areas according to the house attribute data; S2, collecting comprehensive heating data of n heating sub-areas; S3, input the comprehensive heating data into the pre-built LSTM neural network model to analyze the heating load demand of n heating sub-areas in the future t time period; The house attribute data includes house area, house type, house wall insulation rate, house orientation and house floor height; The method for obtaining the thermal insulation rate of the house wall includes: Use an infrared thermal imager to scan the wall of the house to obtain a thermal map of the wall surface temperature; during scanning, ensure that the distance between the infrared thermal imager and the wall is within B meters; while collecting the thermal map of the wall temperature, use a temperature sensor to collect the ambient temperature within a range of L meters around the house wall; through an image processing algorithm, define an area of ​​interest on the thermal map of the wall surface temperature and extract the average temperature within the area of ​​interest; select an area of ​​interest for every 1 square meter of the house wall scanned; repeat the scan C times to obtain sample data of the wall surface temperature; Establish a heat transfer model for the wall based on the material data of the wall material; combine the heat transfer model with the sample data of the wall surface temperature and the ambient temperature, and use the optimization algorithm to inversely solve the wall insulation rate; The heat transfer model is established in the following manner: Check the product manual or test report of the wall material to determine the material data of the wall material; the material data includes density, specific heat capacity and thermal conductivity of the material; A one-dimensional steady-state heat conduction model is established. Using the one-dimensional control volume method, the wall is equivalent to E two-sided control units, each of which contains four boundary surfaces. The energy conservation equation of the two-sided control unit is defined as: ; in, is the thickness of the two-sided control unit; is the thermal conductivity of the material, is the boundary surface area, and are the heat flux densities at the left and right interfaces respectively; Discretize the energy conservation equation to get the heat transfer model ; in, , They are the sample data of wall surface temperature and ambient temperature respectively; is the coefficient matrix, is the heat flow vector; The method of combining the heat transfer model with the wall surface temperature sample data and the ambient temperature and using the optimization algorithm to inversely solve the wall insulation rate includes: Substitute the ambient temperature into the wall heat transfer model and initialize the value range of the wall insulation rate ; exist Select Y insulation rates; construct a corresponding coefficient matrix according to each insulation rate, substitute the coefficient matrix into the heat transfer model, and calculate the wall surface temperature; Define the mean square error between the temperature output by the heat transfer model and the wall surface temperature sample data; Mean square error ;in, Temperature output for heat transfer models; For the Sample data of wall surface temperature, is the total number of wall surface temperature sample data; Calculate the mean square error between the model output temperature and the wall surface temperature sample data; use the genetic algorithm to search for the insulation rate to minimize the mean square error and obtain the final insulation rate value; The method of dividing the heating area into n heating sub-areas includes: The autoregressive integrated moving average algorithm is used to establish a quantitative relationship model between the attribute and the load; the quantitative relationship model is established in the following manner: Extract the optimal relationship data from the house attribute data, use the optimal relationship data as the exogenous variable of the autoregressive integrated moving average algorithm, and use the heating load corresponding to the optimal relationship data as the dependent variable of the autoregressive integrated moving average algorithm; Initialize the hyperparameters of the autoregressive integrated moving average algorithm, bring the exogenous variables and dependent variables into the autoregressive integrated moving average algorithm, and use the maximum likelihood estimation method for training; obtain the constant term, AR coefficient, and MA coefficient of the autoregressive integrated moving average algorithm; and obtain the quantitative relationship model by analyzing the coefficients of each exogenous variable; The house is initially divided into v cubes, and the average value of the heating load in each cube is calculated according to the quantitative relationship model as the average value variable; the world coordinates of the center point of each cube are recorded; the Euclidean distance between the cubes is calculated according to the world coordinates of the center point of each cube; Pearson correlation coefficients were calculated based on the Euclidean distances between cubes and the mean variables; Take all the divided v cubes as sample areas, calculate the Pearson correlation coefficients between the v cubes, and form a v×v symmetric matrix R; the first Line The elements on the column represent cubes and cube Pearson correlation coefficient between ; For each sample area Calculate the average Pearson correlation coefficient between it and other samples in the cluster and its region , and the heterogeneous average Pearson correlation coefficient between it and all samples in the nearest cluster ; Calculate the silhouette coefficient of the sample area ; Calculate the overall silhouette coefficient , gradually reduce K and repeat the calculation ; Records Take the maximum value K as the optimal number of categories; the corresponding clustering result is the best regional division; merge the sample areas divided into the same cluster at this time to obtain the heating sub-area; K at this time is the number of heating sub-areas n; The comprehensive heating data includes heating data and meteorological parameter data; Heating data include pipe network flow, supply and return water temperature, unit start and stop status and fuel consumption; meteorological parameter data include outdoor temperature, humidity, wind speed and direction; Numerical encoding of the start and stop status of the unit and the wind direction; defining the corresponding normal data interval for the heating data, removing abnormal values ​​in the historical heating data according to the normal data interval, i.e., data beyond the normal data interval; and mapping data of different dimensions to the same numerical range; Collect comprehensive heating data of n heating sub-areas over a period of time in history, and collect the heating load corresponding to the comprehensive heating data; construct the data into training data; divide the training data into a training set and a validation set; The training set is output to the LSTM neural network model, the heating comprehensive data is received as the input layer, the heating load is used as the output of the output layer, and the n heating sub-areas correspond to the n nodes of the output layer; during the training process, when the function value of the comprehensive loss function does not decrease during the process of P consecutive iterations, the training is completed, and the trained LSTM neural network model is obtained; The collected comprehensive heating data of n heating sub-areas are input into the trained LSTM neural network model to obtain the heating load demand of the n heating sub-areas in the future t time period.

2. A heating supply monitoring and analysis method based on big data according to claim 1, characterized in that: The optimal relationship data is extracted by obtaining preliminary model variables through correlation analysis; performing lasso regression analysis on the preliminary model variables to obtain the optimal relationship data; Ways to obtain preliminary model variables through correlation analysis include: Initialize a correlation matrix; calculate the correlation coefficients between the elements in the correlation matrix; the elements in the correlation matrix are the property data of the house; Use color shades The correlation heat map between variables is drawn according to the numerical value of ; the correlation heat map consists of m grids; each grid is marked with the value of the correlation coefficient; Identify the darker grids in the correlation heat map, find the variable pairs whose correlation coefficients are higher than the preset threshold, that is, there is multicollinearity between the variable pairs; remove the variables with multicollinearity; after removing the variables, recalculate the correlation coefficient matrix; draw a new heat map to determine whether there are still dark grids; if so, identify and remove the variables with multicollinearity again; until there are no multicollinear variables in the correlation heat map; the variables retained at this time are the preliminary model variables; The calculation formula of the correlation coefficient is: ;in, is the correlation matrix Line Elements of the column, representing variables and The correlation coefficient between For variables and The covariance between For variables The standard deviation of For variables The standard deviation of .

3. A heating monitoring and analysis method based on big data according to claim 2, characterized in that: The method of performing lasso regression analysis on the preliminary model variables to obtain the optimal relationship data includes: A penalty term is added to the loss function of the preset linear regression model to obtain a penalty loss function; the penalty loss function is used to perform penalty processing to obtain the next-step model variable; The formula of the linear regression model is: ,in, The characteristic variables obtained for the target; is the input feature matrix; the feature matrix is ​​the matrix composed of the preliminary model variables; is the regression coefficient, is a random error; The penalty loss function is: ; In the formula, is a hyperparameter of the regularization strength; To represent the regression coefficient in the L1 norm The sum index of ; is the total number of preliminary model variables; Indicates Regression coefficients The L1 norm of The punishment methods include: Set an initial , and obtain a sparse regression coefficient vector; variables whose regression coefficients are penalized to 0 are regarded as non-critical variables; variables whose regression coefficients are not 0 are retained; this is the optimal relationship data.

4. A heating monitoring and analysis method based on big data according to claim 3, characterized in that: The construction method of the LSTM neural network model includes: Initialize the basic architecture of the LSTM neural network model, which includes an input layer, two LSTM units, a Dropout layer, a fully connected layer, a learnable parameter activation function, a comprehensive loss function, an optimizer, a regularizer, and an output layer; the output layer includes n nodes; Each LSTM unit consists of a memory unit and three control gates; the three control gates are input gate, forget gate and output gate; a fully connected layer is added between LSTM units; The Dropout layer randomly discards a certain proportion of nodes from the LSTM neural network model; The regularizer uses L2 norm regularization to control the LSTM neural network model to prevent overfitting; the output layer including n nodes is used to output the heating load demand in the future t time period corresponding to the n heating sub-areas.

5. A heating monitoring and analysis method based on big data according to claim 4, characterized in that: The learnable parameter activation function ;in, For input data; is a learnable slope parameter; is a learnable offset parameter; The comprehensive loss function ;in, is the quantile, with a value between (0, 1); is the real target value; is the lower quantile of the predicted distribution; is the upper quantile of the predicted distribution; is a positive function.

6. A heating supply monitoring and analysis method based on big data according to claim 5, characterized in that: The optimizer adopts an adaptive optimizer that adaptively adjusts the learning rate; The optimization formula of the adaptive optimizer is: ;in, The number of training iterations is LSTM neural network model Layer parameters; The number of training iterations is LSTM neural network model Layer parameters; It is the LSTM neural network model The learning rate of the layer; The number of training iterations is The exponential moving average of the squared gradient of ; is a constant; The number of training iterations is The parameter gradient of like ;but ;in The number of training iterations is The value of the comprehensive loss function; The number of training iterations is The value of the comprehensive loss function; To reduce the parameters.

7. A heating supply monitoring and analysis system based on big data, which is used to implement a heating supply monitoring and analysis method based on big data as claimed in any one of claims 1 to 6, characterized in that: include: A collection and division module is used to collect house attribute data and divide the heating area into n heating sub-areas according to the house attribute data; The heating collection module is used to collect comprehensive heating data of n heating sub-areas; The load demand prediction module is used to input the comprehensive heating data into the pre-built LSTM neural network model to analyze the heating load demand of n heating sub-areas in the future t time period; each module is connected by wired and / or wireless means to realize data transmission between modules.

Citation Information

Patent Citations

  • Heat supply monitoring analysis method, device and equipment based on big data and medium

    CN115964668A

  • Prediction method and system for regional cooling, heating, cooling and heating loads

    CN111932015A

  • Central air conditioner load prediction method and system

    CN116989442A

  • Building area energy system cooling and heating load data driving prediction method with flexibility and interpretability

    CN117114438A

  • Building HVAC system with modular cascaded model

    US20210041127A1