Method for predicting carbon emissions of building operation in cold region based on neural network

Abstract

The neural network-based cold region building operation carbon emission prediction method belongs to the field of cold region building operation carbon emission prediction.The present cold region building carbon emission prediction has the problem of poor accuracy, which is solved by the present application.The power consumption and heat consumption of the current building to be tested are collected, the carbon emission coefficient method is used to calculate the operation carbon emission data of the current building to be tested, the operation carbon emission data is input into the pre-trained long short-term neural network model, and the prediction value of the carbon emission of the building to be tested is obtained.The present application is mainly used for cold region building operation carbon emission prediction.

Description

Technical Field

[0001] This invention belongs to the field of carbon emission prediction for building operations in cold regions. Background Technology

[0002] The challenges posed by global warming are becoming increasingly severe, and carbon emission control has become one of the top priorities for governments and research institutions worldwide. Cold regions refer to the severely cold and cold regions within the thermal zoning. Buildings in cold regions have high heating demands, and prolonged heating is a significant source of carbon emissions from buildings in these climates. Therefore, improving the accuracy of carbon emission predictions is crucial for developing effective emission reduction measures and achieving carbon neutrality. Currently available neural network-based methods for predicting building carbon emissions still have significant limitations in predicting operational carbon emissions from buildings in cold regions, particularly in failing to consider the complexity of time-series carbon data for building operations, including significant seasonal differences in energy consumption patterns during building operation and the lack of differentiation between heating and non-heating cycles. Therefore, the accuracy of carbon emission predictions remains poor. Summary of the Invention

[0003] This invention aims to address the problem of poor accuracy in existing carbon emission prediction methods for buildings in cold regions by proposing a neural network-based method for predicting carbon emissions from building operations in cold regions.

[0004] The present invention provides a method for predicting carbon emissions from building operations in cold regions based on neural networks, comprising: collecting the electricity consumption and heat consumption of the building under test; using the carbon emission coefficient method to calculate the carbon emissions generated by electricity consumption and the carbon emissions generated by heat consumption of the heating system during the current operation of the building; inputting the carbon emission data generated by electricity consumption and the carbon emission data generated by heat consumption into a pre-trained long short-term neural network model to obtain the predicted value of carbon emissions from the building under test.

[0005] The long short-term neural network model is trained using historical power consumption and heat consumption sample data of the building under test. The specific training method is as follows:

[0006] Step 1: Using the carbon emission factor method, calculate the carbon emission data generated by the building's electricity consumption and heating energy consumption respectively using the pre-processed electricity and heat consumption data.

[0007] Step 2: Use the Pearson product-moment correlation coefficient to conduct a correlation analysis between carbon emissions from electricity consumption and meteorological factors, and obtain the correlation coefficient.

[0008] Step 3: Divide the carbon emission data generated by the electricity consumption into a strong correlation group, a medium correlation group, and a weak correlation group according to the correlation coefficient. Use the outdoor weather comfort index to further divide the meter data in the medium correlation group to obtain the comfortable group and the uncomfortable group.

[0009] Step 4: Establish four long-term and short-term neural network models. Input the carbon emission data of the electricity consumption of the strong correlation group, weak correlation group, comfort group, and uncomfortable group into the four long-term and short-term neural network models respectively to predict the carbon emission of electricity consumption and obtain the predicted carbon emission of electricity consumption.

[0010] At the same time, a fifth long-short-term neural network model is established. The carbon emission data generated by heat energy consumption is input into the fifth long-short-term neural network model to predict the carbon emission generated by heating heat energy consumption and obtain the carbon emission results generated by heating heat energy consumption.

[0011] The predicted carbon emissions from electricity consumption and the predicted carbon emissions from heating energy consumption are summed to obtain the final predicted carbon emissions from building operations.

[0012] Furthermore, in this invention, in step two, the carbon emission coefficient method is used to calculate the carbon emission data generated by the building's electricity consumption and the carbon emission data generated by its heating energy consumption, respectively, using the pre-processed electricity and heat consumption data.

[0013] Using the formula:

[0014]

[0015] To calculate Building Operation Carbon Emissions (CE), BFS is the building area of ​​the test building, and CEI is the Building Operation Carbon Emission Intensity.

[0016] The carbon emission intensity of the building operation is:

[0017]

[0018] Among them, EUI is the energy use index for each type of energy, CEF represents the carbon emission factor for each type of energy, and S represents the amount of energy involved in the operation and carbon emissions.

[0019] The carbon emissions generated by electricity consumption are calculated as follows:

[0020]

[0021] It is an energy consumption index for electrical energy. It is a carbon emission factor of electrical energy;

[0022] The carbon emissions from heating energy consumption are calculated as follows:

[0023] .

[0024] It is the energy consumption index for heating. It is a carbon emission factor of heating energy.

[0025] Furthermore, in this invention, in step three, the method for performing correlation analysis between carbon emission data generated from electricity consumption and meteorological factors to obtain the correlation coefficient is as follows:

[0026] The carbon emission data generated by annual electricity consumption is divided into 12 time periods based on months, and a correlation analysis is conducted between the carbon emission data generated by building electricity consumption and meteorological factors in each time period.

[0027] Pearson product-moment correlation coefficients (PCCs):

[0028]

[0029] in It is the covariance of the ordinal variable. The standard deviation is an ordinal variable, where X and Y are the carbon emissions from building electricity consumption and the average temperature, respectively.

[0030] Furthermore, in this invention, the method for dividing the carbon emission data generated by the electricity consumption into a strongly correlated group, a moderately correlated group, and a weakly correlated group based on the correlation coefficient in step four is as follows:

[0031] Based on the absolute value of the Pearson product-moment correlation coefficient (PCCs), the correlation is divided into four levels: strong correlation group, medium correlation group, and weak correlation group. The weak correlation group includes weakly correlated and uncorrelated data.

[0032] The strong correlation group has an absolute correlation coefficient ranging from 0.5 to 1.0, the moderate correlation group has an absolute correlation coefficient ranging from 0.3 to 0.5, and the weak correlation group has an absolute correlation coefficient ranging from 0 to 0.3.

[0033] Furthermore, in this invention, the method for obtaining the comfort group and the discomfort group in step four is as follows:

[0034] The Outdoor Comfort Index (OWCI) was used to further subgroup the mid-correlation group;

[0035]

[0036] Where C is a constant, C ta For the average temperature term, C tmFor the highest temperature term, C h For humidity, C v For wind speed, C r For the radiation item, data with a comfort index ≥ 7.0 are divided into the comfort group, and data with a comfort index < 7.0 are divided into the discomfort group.

[0037] Furthermore, in this invention, in step five, the long short-term neural network model includes an output gate, a forget gate, an input gate, a matrix calculation module, a first multiplier, a first adder, a second multiplier, and a third multiplier;

[0038] The input data is simultaneously output to the output gate, forget gate, input gate, and matrix calculation module;

[0039] The input gate uses an activation function to control the input switch, thereby controlling whether the input data is input to the third multiplier.

[0040] The forget gate uses an activation function to control whether the predicted data from the previous time step is retained; the retained data is then input into the second multiplier.

[0041] The output gate is used to control whether the input data of the current time step and the prediction value of the previous time step are converted into a hidden state, and the data that is not converted into a hidden state is transmitted to the first multiplier.

[0042] The matrix calculation module is used to perform matrix calculations on the input data, obtain the calculation results, and input the calculation results into the third multiplier;

[0043] The formula for calculating the matrix is:

[0044]

[0045] in Represents the cell state at the previous time step. legacy Indicates the output of the Forgot Gate. This indicates the input to the input gate. Indicates cell candidate value

[0046]

[0047] in It is an activation function that restricts the range of candidate values ​​to [−1, 1]. This indicates the hidden state data from the previous time step. This represents the input data, where Indicates input unit The weight matrix, U c The hidden state of the previous time step The weight matrix of the linear transformation, bc It is a deviation value that ensures that the output can be activated even when the input and hidden states are zero.

[0048] The third multiplier performs element-wise matrix multiplication on the data output from the input gate and the calculation result from the matrix calculation module to obtain a set of candidate cell state values; the set of candidate cell state values ​​is then transmitted to the first adder.

[0049] The second multiplier performs element-wise matrix multiplication on the result output by the first adder at the previous time step and the data retained by the forget gate to obtain the product result; the product result is input to the first adder, the first adder performs element-wise matrix addition on the product result and the set of candidate cell state values, and the result of the element-wise addition is input to the second multiplier and the first multiplier;

[0050] The first multiplier performs a matrix element-major multiplication on the data that has not been converted to a hidden state at the output gate and the result of element-wise addition to obtain the cell's next time step state; the cell's next time step state is then output as a predicted value.

[0051] Furthermore, in this invention, the forget gate uses an activation function to control whether the predicted data from the previous time step is retained, according to the following formula:

[0052]

[0053] This represents the output of the forget gate. This represents the activation function, with a value of 0 or 1, where 0 represents complete forgetting and 1 represents complete retention. This indicates that the forget gate hides the state from the previous time step. The linear transformation weight matrix, This represents the input weight matrix of the forget gate. This indicates a deviation from the forget gate.

[0054] Furthermore, in this invention, the input gate uses an activation function to control the input switch, and the formula for controlling the input data is as follows:

[0055]

[0056] This is the activation function, with a value of 0 or 1, where 0 represents off and 1 represents fully on; it controls the opening and closing of the input gate. This represents the input weight matrix of the input gate. Indicates the hidden state of the input gate at the previous time step. The linear transformation weight matrix, Indicates the deviation of the input gate. This indicates the output of the input gate.

[0057] Furthermore, in this invention, the formula for the output gate (1) to perform a partial state transition between the input data of the current time step and the predicted value of the previous time step is as follows:

[0058]

[0059] This indicates the output of the output gate. This represents the input weight matrix of the output gate. This indicates that the input state is hidden from the previous time step. The linear transformation weight matrix, This indicates the deviation of the output gate.

[0060] Furthermore, in this invention, the predicted value is:

[0061]

[0062] in, This represents the cell state at time step t.

[0063] This invention addresses the differences in energy use between heating and non-heating seasons in office buildings in cold regions by proposing a carbon emission prediction model based on an LSTM neural network. This model overcomes the shortcomings of existing technologies that fail to adequately consider seasonal energy consumption patterns in cold regions. In terms of carbon emission prediction, the model simulates building operation carbon emissions in groups, reducing errors caused by data fluctuations. The improved model effectively handles the complex energy consumption characteristics of cold climate regions, especially scenarios with significant seasonal fluctuations, improving the accuracy of carbon emission prediction for building operations in cold regions and contributing to the achievement of regional carbon neutrality goals. Attached Figure Description

[0064] Figure 1 This is a flowchart of the training method for the long short-term neural network model described in this invention;

[0065] Figure 2 This is a schematic diagram of a long short-term neural network model. Detailed Implementation

[0066] Specific implementation method one: Combining Figure 1 and Figure 2 This embodiment describes a neural network-based method for predicting carbon emissions from building operations in cold regions. The method includes: collecting the electricity and heat consumption of the building under test; using the carbon emission coefficient method to calculate the carbon emissions generated by electricity consumption and the carbon emissions generated by heat consumption during the building's current operation; inputting the data on carbon emissions generated by electricity consumption and heat consumption into a pre-trained long short-term neural network model to obtain a predicted value for the building's operational carbon emissions.

[0067] The long short-term neural network model is trained using historical power consumption and heat consumption sample data of the building under test. The specific training method is as follows:

[0068] Step 1: Using the carbon emission factor method, calculate the carbon emission data generated by the building's electricity consumption and the carbon emission data generated by its heating energy consumption using sample data of electricity and heat consumption.

[0069] Step 2: Use the Pearson product-moment correlation coefficient to conduct a correlation analysis between carbon emissions from electricity consumption and meteorological factors, and obtain the correlation coefficient.

[0070] Step 3: Divide the carbon emission data generated by the electricity consumption into a strong correlation group, a medium correlation group, and a weak correlation group according to the correlation coefficient. Use the outdoor weather comfort index to further divide the meter data in the medium correlation group to obtain the comfortable group and the uncomfortable group.

[0071] Step 4: Establish four long-term and short-term neural network models. Input the carbon emission data of the electricity consumption of the strong correlation group, weak correlation group, comfort group, and uncomfortable group into the four long-term and short-term neural network models respectively to predict the carbon emission of electricity consumption and obtain the predicted carbon emission of electricity consumption.

[0072] At the same time, a fifth long-short-term neural network model is established. The carbon emission data generated by heat energy consumption is input into the fifth long-short-term neural network model to predict the carbon emission generated by heating heat energy consumption and obtain the carbon emission results generated by heating heat energy consumption.

[0073] The predicted carbon emissions from electricity consumption and the predicted carbon emissions from heating energy consumption are summed to obtain the final predicted carbon emissions from building operations.

[0074] The historical power consumption and heat consumption sample data of the building to be tested described in this invention are the historical power consumption and heat consumption of the building to be tested for at least one year per hour, and the power consumption and heat consumption have been preprocessed.

[0075] The data collection interval for electricity meter and heat meter data used in this invention should be one hour, and the data time span should be at least one year. The electricity meter and heat meter data should be summarized and organized by item in conjunction with the time series. Missing data can be filled using statistical methods such as linear interpolation, but the continuous time of missing data should not exceed 24 hours.

[0076] The carbon emission data of the building operation to be tested in this invention includes: carbon emission data of the building caused by electricity consumption and carbon emission data of the building caused by heat consumption; during the non-central heating period, the daily building operation carbon is the predicted result of the carbon emission generated by electricity consumption, and the carbon emission data of the building caused by heat consumption is 0; during the central heating period, the daily building operation carbon is the sum of the predicted results of the carbon emission generated by electricity consumption and the carbon emission generated by heat consumption.

[0077] Furthermore, in this invention, in step two, the carbon emission coefficient method is used to calculate the carbon emission data generated by the building's electricity consumption and the carbon emission data generated by its heating energy consumption, respectively, using the pre-processed electricity and heat consumption data.

[0078] Using the formula:

[0079]

[0080] To calculate Building Operation Carbon Emissions (CE), BFS is the building area of ​​the test building, and CEI is the Building Operation Carbon Emission Intensity.

[0081] The carbon emission intensity of the building operation is:

[0082]

[0083] Among them, EUI is the energy use index for each type of energy, CEF represents the carbon emission factor for each type of energy, and S represents the amount of energy involved in the operation and carbon emissions.

[0084] The carbon emissions generated by electricity consumption are calculated as follows:

[0085]

[0086] It is an energy consumption index for electrical energy. It is a carbon emission factor of electrical energy;

[0087] The carbon emissions from heating energy consumption are calculated as follows:

[0088] .

[0089] It is the energy consumption index for heating. It is a carbon emission factor of heating energy.

[0090] Furthermore, in this invention, in step three, the method for performing correlation analysis between carbon emission data generated from electricity consumption and meteorological factors to obtain the correlation coefficient is as follows:

[0091] The carbon emission data generated by annual electricity consumption is divided into 12 time periods based on months, and a correlation analysis is conducted between the carbon emission data generated by building electricity consumption and meteorological factors in each time period.

[0092] Pearson product-moment correlation coefficients (PCCs):

[0093]

[0094] Among them, It is the covariance of the ordinal variable. The standard deviation is an ordinal variable, where X and Y are the carbon emissions from building electricity consumption and the average temperature, respectively.

[0095] Furthermore, in this invention, the method for dividing the carbon emission data generated by the electricity consumption into a strongly correlated group, a moderately correlated group, and a weakly correlated group based on the correlation coefficient in step four is as follows:

[0096] Based on the absolute value of the Pearson product-moment correlation coefficient (PCCs), the correlation is divided into four levels: strong correlation group, medium correlation group, and weak correlation group. The weak correlation group includes weakly correlated and uncorrelated data.

[0097] The strong correlation group has an absolute correlation coefficient ranging from 0.5 to 1.0, the moderate correlation group has an absolute correlation coefficient ranging from 0.3 to 0.5, and the weak correlation group has an absolute correlation coefficient ranging from 0 to 0.3.

[0098] Furthermore, in this invention, the method for obtaining the comfort group and the discomfort group in step four is as follows:

[0099] The Outdoor Comfort Index (OWCI) was used to further subgroup the mid-correlation group;

[0100]

[0101] Where C is a constant, C ta For the average temperature term, C tm For the highest temperature term, C h For humidity, C v For wind speed, C r For the radiation item, data with a comfort index ≥ 7.0 are divided into the comfort group, and data with a comfort index < 7.0 are divided into the discomfort group.

[0102] Furthermore, in this invention, in step five, the long short-term neural network model includes an output gate 1, a forget gate 2, an input gate 3, a matrix calculation module 4, a first multiplier 5, a first adder 6, a second multiplier 7, and a third multiplier 8;

[0103] The input data is simultaneously output to output gate 1, forget gate 2, input gate 3, and matrix calculation module 4;

[0104] The input gate 3 uses an activation function to control the input switch and control whether the input data is input to the third multiplier 8;

[0105] Forget Gate 2 uses an activation function to control whether the predicted data from the previous time step is retained; the retained data is then input into the second multiplier 7.

[0106] Output gate 1 is used to control whether the input data of the current time step and the predicted value of the previous time step are converted into a hidden state, and the data that is not converted into a hidden state is transmitted to the first multiplier 5.

[0107] Matrix calculation module 4 is used to perform matrix calculations on the input data, obtain the calculation results, and input the calculation results to the third multiplier 8;

[0108] The formula for calculating the matrix is:

[0109]

[0110] in Represents the cell state at the previous time step. legacy Indicates the output of the Forgot Gate. This indicates the input to the input gate. Indicates cell candidate value

[0111]

[0112] in It is an activation function that restricts the range of candidate values ​​to [−1, 1]. This indicates the hidden state data from the previous time step. This represents the input data, where Indicates input unit The weight matrix, U c The hidden state of the previous time step The weight matrix of the linear transformation, b c It is a deviation value that ensures that the output can be activated even when the input and hidden states are zero.

[0113] The third multiplier 8 performs element-wise matrix multiplication on the data output from the input gate 3 and the calculation result of the matrix calculation module 4 to obtain a set of candidate cell state values; the set of candidate cell state values ​​is transmitted to the first adder 6.

[0114] The second multiplier 7 performs element-wise matrix multiplication on the result output by the first adder 6 at the previous time step and the data retained by the forget gate 2 to obtain the product result; the product result is input to the first adder 6, the first adder 6 performs element-wise matrix addition on the product result and the set of candidate cell state values, and the result of the element-wise addition is input to the second multiplier 7 and the first multiplier 5;

[0115] The first multiplier 5 performs a matrix element-major multiplication on the data that has not been converted to the hidden state from the output gate 1 and the result of element-wise addition to obtain the cell's next time step state; the cell's next time step state is output as a predicted value.

[0116] Furthermore, in this invention, the forget gate 2 uses an activation function to control whether the predicted data from the previous time step is retained, according to the following formula:

[0117]

[0118] This represents the output of the forget gate. This represents the activation function, with a value of 0 or 1, where 0 represents complete forgetting and 1 represents complete retention. This indicates that the forget gate hides the state from the previous time step. The linear transformation weight matrix, This represents the input weight matrix of the forget gate. This indicates a deviation from the forget gate.

[0119] Furthermore, in this invention, the input gate 3 uses an activation function to control the input switch, and the formula for controlling the input data is as follows:

[0120]

[0121] This is the activation function, with a value of 0 or 1, where 0 represents off and 1 represents fully on; it controls the opening and closing of the input gate. This represents the input weight matrix of the input gate. Indicates the hidden state of the input gate at the previous time step. The linear transformation weight matrix, Indicates the deviation of the input gate. This indicates the output of the input gate.

[0122] Furthermore, in this invention, the formula for output gate 1 to perform a partial state transition between the current time step input data and the predicted value of the previous time step is as follows:

[0123]

[0124] This indicates the output of the output gate. This represents the input weight matrix of the output gate. This indicates that the input state is hidden from the previous time step. The linear transformation weight matrix, This indicates the deviation of the output gate.

[0125] Furthermore, in this invention, the predicted value is:

[0126]

[0127] in, This represents the cell state at time step t.

[0128] This invention decomposes the complex annual fluctuations in building operation carbon emissions into carbon emissions from electricity consumption and carbon emissions from heat consumption based on meteorological characteristics and outdoor weather numerical indices. Furthermore, it decomposes the carbon emission data from electricity consumption into four sets of building operation carbon emission data, simulates them separately, and then combines them. The prediction accuracy is improved compared to directly simulating the data for the whole year.

[0129] In this embodiment, the Long Short-Term Neural Network (LSTM) is currently the most commonly used method for predicting carbon emissions from building operations. However, when faced with long-term series data, it can lead to a decrease in the model's learning ability and a reduction in prediction accuracy.

[0130] Long-term data series often contains multiple periods. For example, carbon emissions data from building operations may fluctuate over fixed periods of a day, a week, or a year. Differences in electricity consumption between day and night, work patterns between weekdays and weekends, and peak carbon emissions in winter and summer are all examples of periods. Training the model directly on complete data requires it to learn multiple patterns, increasing the training difficulty. In this case, the model may not fully learn each period, exhibiting a phenomenon where it only learns the most salient features while ignoring details.

[0131] Therefore, grouping the data reduces data complexity, highlights feature correlations, and avoids noise interference. After each grouping, the model only needs to process one simple pattern, reducing complexity while improving prediction performance and generalization ability.

[0132] Step S1: Collect raw data.

[0133] Carbon emissions from building operations in cold regions mainly include carbon emissions from electricity consumption and carbon emissions from heating energy consumption. Therefore, hourly electricity and heat meter data for test buildings in cold regions should be collected, with a data span of at least one year. The electricity and heat meter data should be summarized and organized by item in conjunction with the time series. Missing data can be filled using statistical methods such as linear interpolation, but the continuous period of missing data should not exceed 24 hours.

[0134] Step S2: Calculate the building operation carbon emission data from the electricity meter data and the heat meter data.

[0135] The carbon emission factor method is used to calculate the carbon emissions generated by electricity consumption and the carbon emissions generated by heat consumption, respectively.

[0136] The carbon emission factor method mentioned above refers to the building operation carbon emission calculation model developed by the Intergovernmental Panel on Climate Change (IPCC). This model combines activity level data with carbon emission factors. The specific building operation carbon emission calculation formula (1) is as follows:

[0137]

[0138] Where CE is building operation carbon emissions, BFS is the building area of ​​the test building, and CEI is building operation carbon intensity.

[0139] The formula (2) for calculating the carbon emission intensity of building operation is as follows:

[0140]

[0141] Where EUI is the energy use index for each energy source, CEF represents the carbon emission factor for each energy source, and S represents the amount of energy involved in operational carbon emissions. The building operation carbon emission intensity is obtained by multiplying the energy use index and carbon emission factor for each energy source and then summing them.

[0142] The formula for calculating carbon emissions from electricity consumption is as follows:

[0143] .

[0144] The formula for calculating carbon emissions from heat energy consumption is as follows:

[0145] .

[0146] The electricity carbon emission factor (CEF) adopts the average carbon dioxide emission factor of electricity in Northeast China published by the National Bureau of Statistics in 2021, with a value of 0.6012 (kgCO2 / kWh).

[0147] The specific model for calculating carbon emissions from building operations is shown in the figure below:

[0148] Step S3: Initial grouping, and correlation analysis using Pearson product-moment correlation coefficient.

[0149] The operations from step three to step six are only applied to the carbon emissions generated by electricity consumption, and no action is taken on the carbon emissions generated by heat consumption.

[0150] Pearson product-moment correlation coefficients (PCCs) were used to perform correlation analysis between carbon emissions data from the building's electricity consumption and meteorological factors.

[0151] The formula for calculating the Pearson product-moment correlation coefficient (PCCs) is as follows:

[0152]

[0153] in It is the covariance of the ordinal variable. It is the standard deviation of an ordinal variable.

[0154] The carbon emissions generated by building electricity consumption throughout the year are divided into 12 time periods based on the month, and a correlation analysis between the carbon emissions generated by building electricity consumption and meteorological factors is conducted in each time period.

[0155] Based on the absolute value of the Pearson product-moment correlation coefficient (PCCs), the correlation is divided into four levels: strong, moderate, weak, and none. The classification criteria are shown in Table 1.

[0156] Table 1. Criteria for Pearson Correlation Coefficient Classification

[0157]

[0158] Step S4: (Group the data according to the index, extract the data groups with obvious characteristic values ​​based on the correlation of meteorological factors, and further group the data with "medium" correlation values, that is, data with some characteristic values ​​but not particularly strong correlation values.)

[0159] Through the correlation calculation in step S3, the carbon emission data from building electricity consumption throughout the year is divided into three groups: G1, G2, and T. G1 represents the period when the carbon emission data from building electricity consumption has a weak or no correlation with meteorological factors, i.e. The range is from 0.0 to 0.3; G2 is the time period where carbon emissions from building electricity consumption are strongly correlated with meteorological factors, i.e. The time period ranges from 0.5 to 1.0; T represents the time period in the correlation between carbon emissions from building electricity consumption and meteorological factors, i.e. The time period is in the range of 0.3 to 0.5.

[0160] In the naming of groups, G represents Group and T represents Transition.

[0161] Step S5: For the T group in the correlation between carbon emissions from building electricity consumption and meteorological factors in Step S4, further refine the grouping using the Outdoor Weather Comfort Index (OWCI). The formula for calculating the Outdoor Weather Comfort Index (OWCI) is as follows:

[0162]

[0163] Where C is a constant, C ta For the average temperature term, C tm For the highest temperature term, C b For humidity, C v For wind speed, C r It is a radiation term.

[0164] The data on carbon emissions from building electricity consumption is in the "medium" range of correlation with meteorological data. This data period often occurs during transitional months, and temperatures fluctuate greatly during the transitional season in cold regions. Therefore, the Outdoor Comfort Index (OWCI) is used for analysis to extract features and perform more detailed grouping.

[0165] The existing Outdoor Comfort Index (OWCI) corresponds to different comfort level classifications. There are four levels: Extremely Comfortable, Comfortable, Incompatible, and Extremely Incompatible. The specific temperature indicators and index values ​​for each level are shown in Table 2.

[0166] Table 2 OWCI Classification Temperature Indicators and Their Index Values

[0167]

[0168] Where T a The average temperature of the day, T max The highest temperature of the day.

[0169] In step four, the outdoor weather comfort index is calculated daily for the time period T in the correlation between building electricity consumption and meteorological factors, and two building operation carbon emission time periods, T1 and T2, are formed based on the results.

[0170] Where T1 refers to the time period with output code 1, and T2 refers to the time period with output code 3. If the output code is 2 or 4, then the time period for that day is the same as the previous day.

[0171] Step S6: Based on steps S3, S4, and S5, the carbon emissions generated by the building's annual electricity consumption can be divided into four time periods: G1, G2, T1, and T. 2；

[0172] After steps S3, S4 and S5, four sets of carbon emission data generated by building electricity consumption can be obtained. These four sets of data cover the whole year and have obvious periodicity and characteristic values.

[0173] Through the above steps, based on meteorological factors and outdoor weather comfort index, the carbon emission data generated by the annual building electricity consumption is decomposed from a complex dataset into four simple datasets, and a Long Short-Term Neural Network (LSTM) model is established for each of the four simple datasets. Simultaneously, a separate LSTM model is established for the carbon emissions generated by thermal energy consumption, and the model is trained separately for this purpose. This leads to a new time series prediction framework, which is composed of the aforementioned five data models.

[0174] The Long Short-Term Neural Network (LSTM) model described above is a variant of the traditional Recurrent Neural Network (RNN). LSTM adds a gating system to the memory unit, namely a forget gate, an input gate, and an output gate. The specific memory unit structure is as follows: Figure 2 As shown;

[0175] In time series forecasting, a "time step" usually refers to the time interval between one observation point and the next observation point in the series data. It defines the sampling frequency and time scale of the time series data. The operation of the current time step is indicated by the subscript t, and the content of the previous time step is indicated by the subscript t-1.

[0176] The input data in the diagram refers to the data at the current time step that is currently input to the LSTM unit. ;

[0177] The forget gate determines whether historical information is retained; its formula is:

[0178]

[0179] in the formula This is the activation function, and its value is usually 0 or 1, where 0 represents complete forgetting and 1 represents complete retention.

[0180] The function of the input gate is to determine which parts of the input data at the current time step are written into the input. Its formula is:

[0181]

[0182] in the formula This is the activation function, used to control the degree to which the input gates open and close.

[0183] The output gate determines which parts of the current time step are output to the hidden state; its formula is:

[0184]

[0185] in the formula Activation function It is used to control the opening and closing of the output gate.

[0186] In the above formula, , , These are the input units of each gate. The weight matrix is ​​transformed in this way, which enables the long short-term neural network to extract complex patterns from the input features. , , The hidden state of the previous time step The weight matrix is ​​a linear transformation of the matrix, which allows long short-term neural networks to consider historical information when deciding on the current state; It is the hidden state, which is the output of the previous time step of the long short-term neural network, and also serves as one of the inputs of the current time step. Its function is to act as a dynamic memory, passing short-term information between time steps. , and This represents the deviation of the door, and its function is to help adjust the door's activation value so that the door's output can be activated even when the input and hidden states are zero.

[0187] In the picture and The multiplication sign inside the square brackets represents element-wise matrix multiplication, and the addition sign represents element-wise matrix addition. The formula for updating the cell state is:

[0188]

[0189] in It is the cell state at the previous time step. Multiply by the Gate of Oblivion The output, It is the input gate With cell candidate value The product of . Candidate cell states. This refers to input data Hidden state at the previous time step Starting from this point, the result is generated after a set of weight calculations, and its generation formula is:

[0190]

[0191] in It is an activation function that restricts the range of candidate values ​​to [−1, 1] to prevent values ​​from being too large or too small.

[0192] The predicted value in the final figure is the final output at the current time step. It also serves as the hidden state for the next time step, and its formula is:

[0193]

[0194] The aforementioned gating system can control the flow of information in the neural network, thereby enabling the processing of short-term and long-term dependencies and effectively solving time-series-based problems. Therefore, a long short-term neural network (LSTM) is used for simulation and prediction.

[0195] Step S7: Train the neural network. The relevant parameters for training the long and short term neural network model were set in advance.

[0196] The input values ​​in a long short-term neural network have three dimensions (m, T). x ,n x A matrix of ).

[0197] Where m is the number of samples in each data set, and T x n is the number of time steps for each sample group. x It represents the number of input features.

[0198] The grouped data was divided into training, validation, and test groups in a ratio of 7:2:1. The Long Short-Term Memory (LSTM) neural network was trained, and training was stopped before overfitting, resulting in the trained neural network model.

[0199] The specific parameter settings for the model are as follows:

[0200] Table 3 Parameter values ​​for long short-term neural network models

[0201]

[0202] Step S8: Calculate building operation carbon emissions; use the trained model to predict carbon emissions from electricity consumption and heating energy consumption. The prediction results overlap during certain time periods, specifically the annual centralized heating season. The resulting building operation carbon predictions are as follows:

[0203] During periods when there is no centralized heating, the predicted carbon emissions from building operations are the carbon emissions generated by electricity consumption.

[0204] During the period of centralized heating, building operation carbon is the sum of the predicted carbon emissions from electricity consumption and the predicted carbon emissions from heating energy consumption.

[0205] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other described embodiments.

Claims

1. A method for predicting carbon emissions from building operations in cold regions based on neural networks, characterized in that, include: The power consumption and heat consumption of the building under test are collected. The carbon emission coefficient method is used to calculate the carbon emissions generated by the power consumption and the carbon emissions generated by the heat consumption of the heating system during the current operation of the building. The carbon emission data generated by the power consumption and the carbon emission data generated by the heat consumption are input into a pre-trained long short-term neural network model to obtain the predicted value of the carbon emissions of the building under test. The long short-term neural network model is trained using historical power consumption and heat consumption sample data of the building under test. The specific training method is as follows: Step 1: Using the carbon emission factor method, calculate the carbon emission data generated by the building's electricity consumption and heating energy consumption respectively using the pre-processed electricity and heat consumption data. Step 2: Use the Pearson product-moment correlation coefficient to conduct a correlation analysis between carbon emissions from electricity consumption and meteorological factors, and obtain the correlation coefficient. The method for obtaining the correlation coefficient by performing correlation analysis between carbon emission data from electricity consumption and meteorological factors is as follows: The carbon emission data generated by annual electricity consumption is divided into 12 time periods based on months, and a correlation analysis is conducted between the carbon emission data generated by building electricity consumption and meteorological factors in each time period. Pearson product-moment correlation coefficients (PCCs): in, It is the covariance of the ordinal variable. The standard deviation is a rank variable, where X and Y are the carbon emissions from building electricity consumption and the average temperature, respectively. Step 3: Divide the carbon emission data generated by the electricity consumption into a strong correlation group, a medium correlation group, and a weak correlation group according to the correlation coefficient. Use the outdoor weather comfort index to further divide the meter data in the medium correlation group to obtain the comfortable group and the uncomfortable group. The method for obtaining the comfort group and discomfort group is as follows: The Outdoor Comfort Index (OWCI) was used to further subgroup the mid-correlation group; Where C is a constant, C ta For the average temperature term, C tm For the highest temperature term, C h For humidity, C v For wind speed, C r For the radiation item, data with a comfort index ≥ 7.0 are divided into the comfort group, and data with a comfort index < 7.0 are divided into the discomfort group; Step 4: Establish four long-term and short-term neural network models. Input the carbon emission data of the electricity consumption of the strong correlation group, weak correlation group, comfort group, and uncomfortable group into the four long-term and short-term neural network models respectively to predict the carbon emission of electricity consumption and obtain the predicted carbon emission of electricity consumption. At the same time, a fifth long-short-term neural network model is established. The carbon emission data generated by heat energy consumption is input into the fifth long-short-term neural network model to predict the carbon emission generated by heating heat energy consumption and obtain the carbon emission results generated by heating heat energy consumption. The predicted carbon emissions from electricity consumption and the predicted carbon emissions from heating energy consumption are summed to obtain the final predicted carbon emissions from building operations.

2. The method for predicting carbon emissions from cold-region building operations based on neural networks according to claim 1, characterized in that, In step one, the carbon emission factor method is used to calculate the carbon emissions from the building's electricity consumption and heating energy consumption, respectively, using the pre-processed electricity and heat consumption data. Using the formula: To calculate Building Operation Carbon Emissions (CE), BFS is the building area of ​​the test building, and CEI is the Building Operation Carbon Emission Intensity. The carbon emission intensity of the building operation is: Among them, EUI is the energy use index for each type of energy, CEF represents the carbon emission factor for each type of energy, and S represents the amount of energy involved in the operation and carbon emissions. The carbon emissions generated by electricity consumption are calculated as follows: It is an energy consumption index for electrical energy. It is a carbon emission factor of electrical energy; The carbon emissions from heating energy consumption are calculated as follows: It is the energy consumption index for heating. It is a carbon emission factor of heating energy.

3. The method for predicting carbon emissions from cold-region building operations based on neural networks according to claim 1, characterized in that, In step three, the method for dividing the carbon emission data generated by the electricity consumption into strongly correlated, moderately correlated, and weakly correlated groups based on the correlation coefficient is as follows: Based on the absolute value of the Pearson product-moment correlation coefficient (PCCs), the correlation is divided into four levels: strong correlation group, medium correlation group, and weak correlation group. The weak correlation group includes weakly correlated and uncorrelated data. The strong correlation group has an absolute correlation coefficient ranging from 0.5 to 1.0, the moderate correlation group has an absolute correlation coefficient ranging from 0.3 to 0.5, and the weak correlation group has an absolute correlation coefficient ranging from 0 to 0.

3.

4. The method for predicting carbon emissions from cold-region building operations based on neural networks according to claim 1, characterized in that, In step five, the long short-term neural network model includes an output gate (1), a forget gate (2), an input gate (3), a matrix calculation module (4), a first multiplier (5), a first adder (6), a second multiplier (7), and a third multiplier (8). The input data is simultaneously output to the output gate (1), the forget gate (2), the input gate (3), and the matrix calculation module (4). The input gate (3) uses an activation function to control the input switch and control whether the input data is input to the third multiplier (8). The forget gate (2) uses an activation function to control whether the predicted data from the previous time step is retained; the retained data is input into the second multiplier (7). The output gate (1) is used to control whether the input data of the current time step and the prediction value of the previous time step are converted into a hidden state, and the data that is not converted into a hidden state is transmitted to the first multiplier (5). The matrix calculation module (4) is used to perform matrix calculations on the input data, obtain the calculation results, and input the calculation results into the third multiplier (8). The third multiplier (8) performs element-wise matrix multiplication on the data output by the input gate (3) and the calculation result of the matrix calculation module (4) to obtain a set of candidate cell state values; and transmits the set of candidate cell state values ​​to the first adder (6). The second multiplier (7) performs a matrix element-wise multiplication on the result output by the first adder (6) at the previous time step and the data retained by the forget gate (2) to obtain the product result; the product result is input to the first adder (6), the first adder (6) performs a matrix element-wise addition on the product result and the set of candidate cell state values, and the result of the element-wise addition is input to the second multiplier (7) and the first multiplier (5); The first multiplier (5) performs matrix element-major multiplication on the data of the output gate (1) that has not been converted to the hidden state and the result of element-wise addition to obtain the cell's next time step state; The cell's next time step state is output as a predicted value.

5. The method for predicting carbon emissions from cold-region building operations based on neural networks according to claim 4, characterized in that, The forget gate (2) uses an activation function to control whether the predicted data from the previous time step is retained. The formula is as follows: This represents the output of the forget gate. This represents the activation function, with a value of 0 or 1, where 0 represents complete forgetting and 1 represents complete retention. This indicates that the forget gate hides the state from the previous time step. The linear transformation weight matrix, This represents the input weight matrix of the forget gate. This indicates a deviation from the forget gate.

6. The method for predicting carbon emissions from cold-region building operations based on neural networks according to claim 5, characterized in that, The input gate (3) uses an activation function to control the input switch, and the formula for controlling the input data is as follows: This is the activation function, with a value of 0 or 1, where 0 represents off and 1 represents fully on; Used to control the opening and closing of the input gate. This represents the input weight matrix of the input gate. Indicates the hidden state of the input gate at the previous time step. The linear transformation weight matrix, Indicates the deviation of the input gate. This indicates the output of the input gate.

7. The method for predicting carbon emissions from cold-region building operations based on neural networks according to claim 6, characterized in that, The formula for the output gate (1) to perform a partial state transition between the input data of the current time step and the predicted value of the previous time step is as follows: This indicates the output of the output gate. This represents the input weight matrix of the output gate. This indicates that the input state is hidden from the previous time step. The linear transformation weight matrix, This indicates the deviation of the output gate.

8. The method for predicting carbon emissions from cold-region building operations based on neural networks according to claim 7, characterized in that, The predicted value is: in, This represents the cell state at time step t.

Images