Cooling and heating electric quantity decomposition method based on kernel density estimation and multiple linear regression

Abstract

The invention discloses a cooling and heating electric quantity decomposition method based on kernel density estimation and multiple linear regression, and the method comprises the steps: S1, constructing a feature tag system based on influence factors; s2, selecting a typical day based on fitting analysis and a kernel density estimation method; S3, predicting basic electric quantity based on multiple linear regression; s4, calculation of cooling and heating electric quantity: the cooling and heating electric quantity is calculated in two parts of not adopting an electricity limiting measure and adopting an electricity limiting measure; s5, result verification based on a Pearson's correlation coefficient method; according to the cooling and heating electric quantity decomposition method based on kernel density estimation and multiple linear regression, the influence of power limiting measures on cooling and heating electric quantity calculation during summer and winter is fully considered, the accuracy of a calculation result is improved, and data support is provided for subsequent electric quantity prediction and power supply guarantee.

Description

Technical Field

[0001] The invention belongs to the technical field of cooling and heating electricity consumption decomposition, and relates to a cooling and heating electricity consumption decomposition method based on kernel density estimation and multivariate linear regression. Background Art

[0002] In recent years, with the improvement of residents' living standards, the number of air conditioners has increased, and the amount of electricity used for cooling and heating has shown a rapid growth trend, and its share in the electricity structure has gradually increased. At the same time, affected by extreme weather, the amount of electricity used for cooling and heating has shown greater volatility. For example, during the summer of 2022, extreme high temperatures were rare in history, air-conditioning equipment continued to operate at a high level, water inflows during the flood season were extremely dry, hydropower generation capacity dropped significantly, and the balance of electricity and power was difficult. The power supply has undergone unprecedented challenges. In addition, the fluctuation of electricity consumption in industrial production is not large, but the electricity consumption of residents is greatly affected by the temperature, which drives the operation of the entire power grid to show a very distinct peak-to-valley difference.

[0003] In the prior art, a cooling and heating electricity calculation model is often constructed based on the benchmark electricity comparison method, with the typical daily electricity in spring and autumn used as the benchmark electricity for non-cooling and heating in the current year. The difference method and trend extrapolation method are used to calculate the natural growth slope, and the natural growth adjustment coefficient is manually set to correct the deviation between the summer and winter periods and the typical basic electricity in spring and autumn. This method subtracts the summer electricity from the basic electricity, and believes that the difference is the cooling electricity affected by temperature in summer. This method has strong applicability, few data sample requirements, and is simple to implement. However, the typical day screening and adjustment coefficient in this method mainly rely on human analysis and intervention, which increases the workload of business personnel, and does not consider the impact of power restriction measures and accumulated temperature effects, which is insufficient to support the accurate calculation of cooling and heating electricity.

[0004] Therefore, it is urgent to carry out research on the optimization of the cooling and heating electricity calculation model, to realize the automatic selection of typical days, to optimize the basic electricity calculation model, and to correct the cooling and heating electricity calculation model while considering the power restriction measures and the accumulated temperature effect, so as to improve the accuracy of cooling and heating electricity calculation and effectively support the normalization of cooling and heating electricity analysis work. Summary of the invention

[0005] In order to optimize the cooling and heating electricity calculation model, the present invention corrects the cooling and heating electricity calculation model considering the power restriction measures and the accumulated temperature effect, realizes automatic selection of typical days, and automatic calculation of adjustment coefficients, thereby improving the accuracy of cooling and heating electricity calculation.

[0006] The technical solution of the present invention is to provide a cooling and heating electricity decomposition method based on kernel density estimation and multivariate linear regression. The key is that the above-mentioned electricity decomposition method includes the steps of:

[0007] S1. Construct a feature label system based on influencing factors;

[0008] S2. Select typical days based on fitting analysis and kernel density estimation:

[0009] S3, predict basic electricity based on multiple linear regression;

[0010] S4. Calculation of cooling and heating electricity: The cooling and heating electricity is divided into two parts: the one without power restriction and the one with power restriction:

[0011] S4-1. For the part where no power restriction measures are taken, the formula Q AC =Q D -Q B Calculate, where Q D is the actual daily power consumption, Q B is the predicted basic power, Q AC Electricity consumption for cooling and heating;

[0012] S4-2. For the part where power restriction measures are taken, the weather and accumulated temperature effects of the day are taken into consideration. Based on the historical data and characteristic data of cooling and heating power without power restriction measures, the multivariate linear regression method is used to predict the cooling and heating power with power restriction measures to obtain the cooling and heating power forecast value, and the forecast value is used to replace the difference value, which is the difference between the actual daily power and the predicted basic power;

[0013] S5. Result verification based on Pearson correlation coefficient method.

[0014] Furthermore, the above step S4-2 specifically includes:

[0015] ① Based on multiple linear regression, the cooling and heating electricity consumption on power-cut-off dates is predicted and a model is obtained;

[0016] ② Considering the accumulated temperature effect, the model needs to be corrected. It is necessary to confirm the critical value of the accumulated temperature effect and the cumulative number of days of the accumulated temperature effect, obtain the temperature correction model of the accumulated temperature effect, substitute the critical value of the accumulated temperature and the correlation coefficient into the temperature correction model of the accumulated temperature effect to obtain the maximum temperature and minimum temperature correction formula;

[0017] Specifically, the formula of the above-mentioned accumulated temperature effect temperature correction model is:

[0018] T'=T+c1(T1-T0)+c2(T2-T0), where T is the temperature of the day, T1 is the temperature of the day before the air conditioner is started, T2 is the temperature of the two days before the air conditioner is started, and c1 is the temperature coefficient of the day before the air conditioner is started c2 is the temperature coefficient of the first two days after the air conditioner is started r1 is the correlation coefficient between the temperature and the cooling and heating electricity of the previous day, r2 is the correlation coefficient between the temperature and the cooling and heating electricity of the previous two days, and r1 and r2 are Pearson correlation coefficients, which are used to describe the degree of linear correlation between temperature and cooling and heating electricity.

[0019] The above-mentioned S2 step specifically includes:

[0020] S2-1. Calculate the perceived temperature using the formula:

[0021] AT = 1.07T + 0.2e-0.65V-2.7, where AT is the perceived temperature, T is the air temperature, e is the water vapor pressure, V is the wind speed, and RH is the relative humidity;

[0022] The calculation formula for the above water vapor pressure e is:

[0023] S2-2, using the formula: Q D =a1·AT 2 +a2·AT+a3, fitting the changing relationship between the annual electricity consumption and the above-mentioned body temperature, and obtaining a U-shaped trend graph;

[0024] S2-3. Obtain the slope Q′ by taking the derivative D =2b1AT+b2, where Q D is the daily electricity consumption, a and b are fitting coefficients, and AT is the perceived temperature;

[0025] S2-4. According to the obtained U-shaped trend graph and slope, the body temperature range corresponding to the bottom of the U-shaped trend graph is selected, and the date corresponding to the selected body temperature range is taken as a typical day. Outliers need to be removed from the above typical days.

[0026] In the above-mentioned S2-4 steps, the elimination of outliers is to first use the kernel density estimation method to statistically calculate the reasonable range of electricity; then calculate the corresponding cumulative distribution function to describe the probability that the random variable is less than or equal to a certain value; finally, set the confidence interval to 90%, eliminate the outliers, and the remaining dates are the typical days.

[0027] Specifically, the above kernel density estimation method is based on the formula Calculate the probability density function for estimating the variable, where Probability density function, n is the number of estimated samples, h is the optimal bandwidth selected by grid search method, K(·) kernel function, and x is the sample data.

[0028] The above step S3-2 specifically includes: finding the difference sequence, finding the difference between two levels, finding the correlation coefficient, and selecting relevant features according to the correlation degree:

[0029] The difference sequence mentioned above is to subtract the corresponding number of each comparison sequence from the reference sequence after initialization. The specific formula is: Δ i (k)=|X0(k)-X i (k)|; i = 1, 2, 3 ... n,

[0030] In the formula, Δi (k) is the absolute value of the difference between the comparison sequence and the reference sequence, X0(k) is the reference sequence, and X i (k) is a comparison sequence;

[0031] The above two-level difference is to find the maximum range and the minimum range. The specific formula is:

[0032]

[0033] Where M 极差 is the maximum range, m 极差 is the minimum range;

[0034] The above-mentioned correlation coefficient is to find the average value of the correlation degree between the comparison sequence and the reference sequence at each point. The specific formula is:

[0035] In the formula, ξ i (k) is the correlation value between the comparison sequence and the reference sequence at each point, Δ i (k) is the absolute value of the difference between the comparison sequence and the reference sequence, ρ is the resolution coefficient, which is 0.5, M is the maximum range, and m is the minimum range;

[0036] The above-mentioned selection of relevant features according to the correlation degree is: sorting the characteristic factors according to the size of the above-mentioned correlation coefficient, obtaining the correlation degree of the characteristic factors to the line loss, selecting relevant features, and eliminating irrelevant features.

[0037] The above S3 step specifically includes:

[0038] S3-1, feature data processing: collecting the typical daily feature data, and converting non-digital feature data labels into digital feature values;

[0039] S3-2. Feature selection based on grey correlation analysis method: Calculate the correlation between basic electricity and feature data using the above grey correlation analysis method, and remove feature values ​​with low correlation;

[0040] S3-3. Predict basic electricity consumption based on multiple linear regression.

[0041] Among them, the above step S3-3 is to substitute the characteristic value processed by step S3-2 into the multivariate linear regression model to obtain the fitting result of the typical daily electricity consumption and the characteristic value. The specific formula is:

[0042]

[0043] In the formula, the n-order polynomial has a0 to a n The n+1 unknown fitting coefficients, is the predicted value and x is the feature value.

[0044] Finally, the above-mentioned S5 step uses the Pearson correlation coefficient method to calculate the correlation coefficient between the comparison benchmark method, the basic electricity prediction method and the cooling and heating electricity calculation results after correction considering power restriction measures and the temperature. If the correlation coefficient after correction considering power restriction measures is higher than the correlation coefficient obtained by the comparison benchmark value method and the basic electricity prediction method, the process ends; if the correlation coefficient after correction considering power restriction measures is lower than the correlation coefficient obtained by the comparison benchmark value method and the basic electricity prediction method, then start again from step S1 and build the feature label system again.

[0045] The beneficial effects of the present invention are: with the goal of improving the calculation accuracy of cooling and heating electricity and reducing human intervention, this paper carries out research on the optimization of the cooling and heating electricity calculation model, analyzes the factors affecting electricity, introduces the calculation formula of perceived temperature, and uses polynomial regression and kernel density estimation to select typical days; based on the results of typical day selection, multivariate linear prediction is used to predict the basic electricity consumption of the air-conditioning season, and the influence of power restriction measures and accumulated temperature effect is considered to correct the cooling and heating electricity prediction model.

[0046] The present invention decomposes the cooling and heating electricity into two parts: no power restriction measures and power restriction measures. When calculating the part where power restriction measures are taken, the influence of the accumulated temperature effect on the cooling and heating electricity is considered. When the temperature is at the same temperature level, there is no obvious increase or decrease in temperature. The relationship between the highest and lowest temperatures and the change in electricity is fitted by polynomials to determine the critical values ​​of the highest and lowest temperatures. The Pearson correlation coefficient method is used to calculate the correlation between the highest and lowest temperatures and the cooling and heating electricity in the previous few days. Analysis shows that the meteorological conditions in the previous two days have a strong correlation with the cooling and heating electricity. Based on the research results of the critical value and the maximum cumulative number of days, a temperature correction model for the accumulated temperature effect is proposed, and the temperature model is verified by the correlation coefficient method. The verification results show that the corrected temperature is more suitable as a characteristic indicator. In the embodiment of the present invention, the cooling electricity in Chongqing in mid-to-late August 2022 is calculated, and the optimization model for the calculation of cooling and heating electricity considering the accumulated temperature effect is verified by the correlation coefficient. Analysis shows that the optimization model for the calculation of cooling and heating electricity considering the accumulated temperature effect improves the accuracy of the calculation of cooling and heating electricity.

[0047] In summary, the cooling and heating electricity decomposition method based on kernel density estimation and multivariate linear regression in the present invention fully considers the impact of power restriction measures to carry out cooling and heating electricity calculation during summer and winter, improves the accuracy of calculation results, and provides data support for subsequent electricity forecasting and power supply guarantee. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is the process of the method for decomposing the electricity consumption for cooling and heating of the present invention.

[0049] Figure 2 This is the fitting relationship diagram between Chongqing electricity consumption and perceived temperature without removing abnormal points.

[0050] Figure 3 This is the fitting relationship diagram between Chongqing electricity consumption and perceived temperature without removing abnormal points.

[0051] Figure 4 Calculation results of daily electricity consumption range in Chongqing.

[0052] Figure 5 Fitting graph of daily electricity consumption and maximum temperature in 2022.

[0053] Figure 6 Fitting graph of daily electricity consumption and minimum temperature in 2022.

[0054] Figure 7 Substitute the pilot unit data to decompose the cooling and heating electricity consumption. DETAILED DESCRIPTION

[0055] The cooling electricity in the present invention refers to the electricity consumption of cooling equipment such as air conditioners and fans during the summer; the heating electricity refers to the electricity consumption of heating equipment such as air conditioners and electric stoves during the winter; the decomposition of cooling and heating electricity means that since there is no metering equipment to collect the electricity consumption of cooling and heating equipment, an algorithm is used to separate the cooling and heating electricity.

[0056] The present invention takes Chongqing city in 2022 as an example to explain the cooling and heating power decomposition method of the present invention. Since the embodiment is aimed at summer, the cooling and heating power is the cooling power.

[0057] S1. Construct a feature label system based on influencing factors:

[0058] During the peak summer and winter periods, weather, accumulated temperature effect, day type, economy, and power restriction measures will all affect electricity consumption. The analysis results are based on the above influencing factors.

[0059] The constructed feature label system is shown in Table 1:

[0060] Table 1: Characteristic indicator system

[0061]

[0062] S2. Select typical days based on fitting analysis and kernel density estimation:

[0063] The electricity and meteorological data for Chongqing from January to December 2022 are collected through the online power grid, data middle platform, and meteorological platform, mainly including the daily maximum temperature, humidity, wind speed, and daily electricity consumption.

[0064] S2-1. Calculate the body temperature:

[0065] The formula AT=1.07T+0.2e-0.65V-2.7 is used, where AT is the perceived temperature, T is the air temperature, e is the water vapor pressure, V is the wind speed, and RH is the relative humidity;

[0066] The calculation formula of water vapor pressure e is:

[0067] S2-2, using the formula: Q D =a1·AT 2 +a2·AT+a3, fitting the changing relationship between annual electricity consumption and perceived temperature, and obtaining a U-shaped trend graph.

[0068] S2-3. Obtain the slope Q′ by taking the derivative D =2b1AT+b2, where Q D is the daily electricity consumption, a and b are the fitting coefficients, and AT is the perceived temperature.

[0069] S2-4. According to the obtained U-shaped trend graph and slope, the body temperature range corresponding to the bottom of the U-shaped trend graph is selected, and the date corresponding to the selected body temperature range is taken as a typical day. Outliers need to be removed from the obtained typical days.

[0070] This is because during January and February, the electricity consumption is relatively high due to some electric heating during the winter. During March to May, the electricity consumption is relatively low and is basically unaffected by the temperature. From June to September, the electricity consumption is the highest due to the high temperature during the summer. The fitting result presents a U-shape, and the electricity consumption corresponding to the bottom of the U-shape is not affected by the temperature and can be used as the electricity consumption on a typical day.

[0071] To remove outliers, we need to remove the influence of typical dates during Qingming Festival, May Day, Dragon Boat Festival, National Day, Mid-Autumn Festival and Spring Festival. The specific process is:

[0072] (1) Use the kernel density estimation method to calculate the reasonable range of electricity, according to the formula In the formula Probability density function, n is the number of estimated samples, h is the optimal bandwidth selected by grid search method, K(·) kernel function, and x is the sample data.

[0073] (2) Calculate the corresponding cumulative distribution function to describe the probability that the random variable is less than or equal to a certain value;

[0074] (3) Finally, the confidence interval is set to 90%, outliers are removed, and the remaining dates are the typical days.

[0075] In this embodiment, the fitting relationship between Chongqing power consumption and body temperature without removing abnormal points is as follows: Figure 2 As shown, R 2 = 0.7804, the fitting relationship between Chongqing power consumption and body temperature after removing the abnormal points is as follows Figure 3As shown, R 2 =0.8391.

[0076] According to the above steps, calculate the slope of the quadratic fitting result, represented by k, that is, the change of power as the perceived temperature increases. D =a1·AT 2 +a2·AT+a3 fitting the changing relationship between power consumption and body temperature, the quadratic fitting function formula is y=31.723x 2 -1361.2x+40108, so the slope k=2×31.723x-1361.2, and then the relationship between the power consumption and the perceived temperature is obtained, and a U-shaped trend graph is obtained. According to the obtained U-shaped trend graph and the slope calculated by statistical data, the date corresponding to the perceived temperature of Chongqing City of 15℃-28℃ is selected as a typical day.

[0077] The probability density analysis is performed by excluding the electricity values ​​during the winter and summer periods. The probability density distribution and the cumulative distribution function diagram are shown in Figure 4. The horizontal axis corresponding to the maximum value is 26270.62, and the electricity value range is [24094.491, 28554.023] when the confidence interval is 90%.

[0078] Therefore, the dates corresponding to the screened body temperature of 15℃-28℃ and the battery level in the range of [24094.491, 28554.023] are typical days.

[0079] S3. Predict basic power based on multiple linear regression:

[0080] S3-1. Feature data processing:

[0081] Collect characteristic data on a typical day, including GDP growth rate, day type, and weather type; convert non-numeric characteristic data labels into numeric characteristic values, such as converting day type labels and weather type labels into numbers, with Monday to Sunday corresponding to numbers 1-7 respectively, and Monday to Friday corresponding to 1, Saturday corresponding to 2, and Sunday corresponding to 3 for working days / rest days.

[0082] S3-2. Feature selection based on grey correlation analysis method: The grey correlation analysis method is used to calculate the correlation between the basic electricity quantity and the characteristic data, and the characteristic values ​​with low correlation are eliminated.

[0083] (1) Find the difference sequence

[0084] The reference sequence after initialization is subtracted from the corresponding number of each comparison sequence, and so on, paying attention to the absolute value.

[0085] Δ i (k)=|X0(k)-X i (k)|; i = 1, 2, 3Λn

[0086] Where Δ i (k) is the absolute value of the difference between the comparison sequence and the reference sequence, X0(k) is the reference sequence, and X i (k) is the comparison sequence.

[0087] (2) Find the difference between the two extremes

[0088]

[0089] Where M 极差 is the maximum range, m 极差 is the minimum range.

[0090] (3) Calculate the correlation coefficient

[0091]

[0092] Where ρ is the resolution coefficient, which is generally taken as 0.5, M is the maximum range, and m is the minimum range;

[0093] Since the correlation coefficient is the correlation degree value between the comparison sequence and the reference sequence at each point, there is more than one. Therefore, the correlation coefficient of each point is averaged as a quantitative representation of the correlation degree between the comparison sequence and the reference sequence.

[0094]

[0095] The characteristic factors are sorted according to the degree of correlation, the correlation degree of the characteristic factors to the line loss is obtained, the relevant features are selected, and the irrelevant features are eliminated.

[0096] S3-3. Prediction of basic electricity quantity based on multivariate linear regression:

[0097] Substitute the eigenvalues ​​processed in step S3-2 into the multiple linear regression model to obtain the fitting results of the typical daily electricity consumption and the eigenvalues. The specific formula is:

[0098]

[0099] In the formula, the n-order polynomial has a0 to a n The n+1 unknown fitting coefficients, is the predicted value (i.e. Q in step 4-1 B ), x is the characteristic value, including GDP growth rate, day type, and meteorological type.

[0100] In this embodiment, the obtained fitting formula is:

[0101]

[0102] In the formula, represents the basic electricity forecast value, x1 represents the quarterly GDP growth rate, x2 represents the day type (Monday to Sunday), x3 represents working days / rest days, and x4 represents the weather type.

[0103] For a comparison of the actual and predicted values ​​of basic electricity, see Figure 5 .

[0104] S4. Calculation of cooling and heating electricity:

[0105] S4-1. Calculation for the part where power restriction measures are not taken:

[0106] Using the formula Q AC =Q D -Q B Calculate, where Q D is the actual daily power consumption, Q B is the predicted basic power, Q AC The power consumption for cooling and heating. The power consumption for cooling and heating without power restriction measures, i.e. the power consumption for cooling, can be found in Figure 6 .

[0107] S4-2. Calculation for power restriction measures:

[0108] Taking into account the cumulative effects of the day's meteorological conditions and accumulated temperature, based on the historical data and characteristic data of cooling and heating electricity consumption without power rationing measures, the multivariate linear regression method is used to predict the cooling and heating electricity consumption with power rationing measures to obtain the cooling and heating electricity prediction value, and the prediction value is used instead of the difference, which is the difference between the actual daily electricity consumption and the predicted basic electricity consumption.

[0109] (1) Prediction of cooling power consumption on power-limited dates based on multivariate linear regression:

[0110] Taking July-August 2022 as an example, the temperature continued to be high in mid-to-late August, but the cooling electricity consumption was relatively low. In fact, power-limiting measures such as demand response and orderly power consumption were taken, which led to the basic electricity forecast value being higher than the actual value, making the cooling electricity calculation result low. Therefore, the calculation result needs to be corrected and the predicted value used instead of the calculated value.

[0111] The prediction algorithm uses multiple linear regression, and the characteristic values ​​include GDP growth rate, day type, working day / rest day, weather type, maximum temperature, minimum temperature, humidity, wind speed, and perceived temperature.

[0112] The fitting formula is:

[0113]

[0114] In the formula, represents the predicted value, x1...x9, respectively represent GDP growth rate, day type, working day / rest day, weather type, maximum temperature, minimum temperature, humidity, wind speed, and body temperature. In this formula, R 2 = 0.925. After replacing the abnormal calculated value with the predicted value, the corrected cooling and heating power calculation result is obtained.

[0115] (2) Considering the accumulated temperature effect, the revised model is modified:

[0116] Since summer load is greatly affected by temperature factors, continuous high temperature weather will cause accumulated temperature effect, resulting in a certain degree of unconventional increase in cooling power. In order to improve the accuracy of cooling power calculation, the impact of accumulated temperature effect on cooling power is analyzed. Taking the temperature of the previous few days into consideration, a characteristic indicator system is constructed, typical features are selected based on the Pearson correlation coefficient method, and the calculation model is optimized.

[0117] In this embodiment, the critical point of accumulated temperature is the temperature at which the air conditioning load is turned on. In order to determine the critical point of the temperature of the accumulated heat effect in summer, the daily maximum temperature, the daily minimum temperature and the daily electricity consumption are fitted respectively, and the influence of holidays is eliminated. The increase in electricity consumption for every 1°C increase is calculated. Since the basic production electricity is not affected by temperature, when the electricity consumption increases with the increase in temperature, it means that the air conditioning load is turned on. The first temperature corresponding to the positive growth is the critical point of accumulated temperature.

[0118] (2-1) Determination of critical value of accumulated temperature effect

[0119] The polynomial regression method is used to fit the daily electricity consumption in Chongqing in 2022 with the maximum and minimum temperatures respectively. The data of one year are used for fitting, and the slopes corresponding to different temperatures are calculated. The slope calculation results are shown in Table 2. The analysis shows that the critical value of the maximum temperature is 24℃ and the critical value of the minimum temperature is 17℃.

[0120] Table 2: Maximum temperature, minimum temperature power growth rate (partial results)

[0121] Temperature / ℃ Maximum temperature power growth rate ℃ / 10,000 kWh Minimum temperature power growth rate ℃ / 10,000 kWh 11 -249.0394 -564.1266 12 -248.7592 -565.0548 13 -264.1258 -525.6282 14 -287.7976 -443.8284 15 -313.225 -319.365 16 -334.6504 -153.6756 17 -347.1082 50.0742 18 -346.4248 286.9908 19 -329.2186 550.4526 20 -292.9 832.11 21 -235.6714 1121.8854 22 -156.5272 1407.9732 23 -55.2538 1676.8398 24 67.5704 1913.2236 25 210.575 2100.135 26 371.5976 2218.8564

[0122] (2-2) Determination of the cumulative number of days of accumulated temperature effect

[0123] The duration of high temperature also has a certain impact on the intensity of the cumulative effect. If too many cumulative days are considered, the intensity of the cumulative effect will no longer be sensitive to the duration of high temperature; if too few cumulative days are considered, it will not be enough to reflect the intensity of the cumulative effect.

[0124] In this embodiment, the Pearson correlation coefficient is used to describe the linear correlation between temperature and cooling and heating power. The specific formula is:

[0125]

[0126] Where r is the correlation coefficient, X is the daily electricity consumption, and Y is the highest and lowest temperatures.

[0127] Substitute the cooling power consumption in Chongqing in July and August 2022 and the highest and lowest temperatures in the previous three days into the above formula. Due to the power restriction measures in mid-to-late August, the cooling power consumption is replaced by the predicted value. To ensure the accuracy of the analysis results, the data in mid-to-late August are excluded for correlation analysis, and the relationship coefficients are shown in Table 3.

[0128] Table 3: Correlation coefficient between cooling power consumption and temperature in the previous few days

[0129] Weather in the past few days Correlation coefficient Maximum temperature 0.896 Minimum temperature 0.908 Previous day's maximum temperature 0.847 Previous day's minimum temperature 0.881 Maximum temperature in the previous two days 0.65 Lowest temperature in the previous two days 0.698 Maximum temperature in the previous three days 0.526 Lowest temperature in the previous three days 0.544

[0130] From the results in Table 3, it can be seen that the two days before the accumulated temperature effect (i.e. the two days before the air conditioner is started) have a strong correlation with the cooling and heating electricity consumption.

[0131] (2-3) The formula of the temperature correction model of the accumulated temperature effect is obtained, which is:

[0132] T' = T + c1 (T1 - T0) + c2 (T2 - T0),

[0133] In the formula, T is the temperature of the day, T1 is the temperature of the day before the air conditioner is started, T2 is the temperature of the two days before the air conditioner is started, and c1 is the temperature coefficient of the day before the air conditioner is started c2 is the temperature coefficient of the first two days after the air conditioner is started r1 is the correlation coefficient between temperature and cooling electricity of the previous day, r2 is the correlation coefficient between temperature and cooling electricity of the previous two days, r1 and r2 are Pearson correlation coefficients, which are used to describe the linear correlation between temperature and cooling and heating electricity.

[0134] Substituting the critical value and correlation coefficient of accumulated temperature in Chongqing urban area into the temperature correction model of accumulated temperature effect, the correction formulas for the maximum and minimum temperatures are obtained:

[0135] Maximum temperature correction formula:

[0136]

[0137] Minimum temperature correction formula:

[0138]

[0139] S5. Result verification based on Pearson correlation coefficient method.

[0140] The Pearson correlation coefficient method is used to calculate the correlation coefficient between the temperature and the calculation results of the cooling and heating electricity after the comparison benchmark value method, the basic electricity prediction method and the correction of the power restriction measures. If the correction correlation coefficient after considering the power restriction measures is higher than the correlation coefficient obtained by the comparison benchmark value method and the basic electricity prediction method, the process ends; if the correction correlation coefficient after considering the power restriction measures is lower than the correlation coefficient obtained by the comparison benchmark value method and the basic electricity prediction method, then start again from step S1 and reconstruct the feature label system.

[0141] In this embodiment, the correlation coefficients between the temperature and the cooling power calculation results after the comparison of the benchmark value method, the basic power prediction method, and the power restriction measures are shown in Table 4.

[0142] Table 4: Comparison of correlation coefficients of different schemes

[0143]

[0144] From the results in Table 4, it can be seen that the correlation coefficient after considering the power restriction measures is higher than that of other schemes, which meets the expected expectations. The results are shown in Table 4. Figure 7 .

Claims

1. A cooling and heating electricity decomposition method based on kernel density estimation and multivariate linear regression is characterized by: The electricity decomposition method comprises the steps of: S1. Construct a feature label system based on influencing factors; S2. Select typical days based on fitting analysis and kernel density estimation: S3, predict basic electricity based on multiple linear regression; S4. Calculation of cooling and heating electricity: The cooling and heating electricity is divided into two parts: the one without power restriction and the one with power restriction: S4-1. For the part where no power restriction measures are taken, the formula Q AC =Q D -Q B Calculate, where Q D is the actual daily power consumption, Q B is the predicted basic power, Q AC Electricity consumption for cooling and heating; S4-2. For the part where power restriction measures are taken, the weather and accumulated temperature effects of the day are taken into consideration. Based on the historical data and characteristic data of cooling and heating power without power restriction measures, the cooling and heating power with power restriction measures is predicted by using the multivariate linear regression method to obtain the cooling and heating power prediction value, and the prediction value is used to replace the difference, which is the difference between the actual daily power and the predicted basic power; S5. Result verification based on Pearson correlation coefficient method.

2. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 1 is characterized in that: Step S4-2 specifically includes: ① Based on multiple linear regression, the cooling and heating electricity consumption on power-cut-off dates is predicted and a model is obtained; ② Considering the accumulated temperature effect, the model needs to be modified. It is necessary to confirm the critical value of the accumulated temperature effect and the cumulative number of days of the accumulated temperature effect to obtain the temperature correction model of the accumulated temperature effect. Substitute the critical value of the accumulated temperature and the correlation coefficient into the temperature correction model of the accumulated temperature effect to obtain the maximum temperature and minimum temperature correction formulas.

3. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 2 is characterized in that: The formula of the temperature correction model of the accumulated temperature effect is: T'=T+c1(T1-T0)+c2(T2-T0), where T is the temperature of the day, T1 is the temperature of the day before the air conditioner is started, T2 is the temperature of the two days before the air conditioner is started, and c1 is the temperature coefficient of the day before the air conditioner is started c2 is the temperature coefficient of the first two days after the air conditioner is started r1 is the correlation coefficient between the temperature and the cooling and heating electricity of the previous day, r2 is the correlation coefficient between the temperature and the cooling and heating electricity of the previous two days, and r1 and r2 are Pearson correlation coefficients, which are used to describe the degree of linear correlation between temperature and cooling and heating electricity.

4. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 1 is characterized in that: The S2 step specifically includes: S2-1. Calculate the perceived temperature using the formula: AT = 1.07T + 0.2e-0.65V-2.7, where AT is the perceived temperature, T is the air temperature, e is the water vapor pressure, V is the wind speed, and RH is the relative humidity; The calculation formula of the water vapor pressure e is: S2-2, using the formula: Q D =a1·AT 2 +a2 AT+a 3 , fitting the changing relationship between the annual electricity consumption and the perceived temperature to obtain a U-shaped trend graph; S2-3. Obtain the slope Q′ by taking the derivative D =2b1AT+b2, where Q D is the daily electricity consumption, a and b are fitting coefficients, and AT is the perceived temperature; S2-4. According to the obtained U-shaped trend graph and slope, the temperature range corresponding to the bottom of the U-shaped trend graph is selected, and the date corresponding to the selected temperature range is taken as a typical day. Outliers need to be removed from the typical day.

5. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 4 is characterized in that: The elimination of outliers in step S2-4 is to first use the kernel density estimation method to count the reasonable range of electricity; then calculate the corresponding cumulative distribution function to describe the probability that the random variable is less than or equal to a certain value; finally set the confidence interval to 90%, eliminate the outliers, and the remaining dates are the typical days.

6. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 5 is characterized in that: The kernel density estimation method is based on the formula Calculate the probability density function for estimating the variable, where Probability density function, n is the number of estimated samples, h is the optimal bandwidth selected by grid search method, K(·) kernel function, and x is the sample data.

7. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 1 is characterized in that: The S3 step specifically includes: S3-1, feature data processing: collecting the typical daily feature data, and converting non-digital feature data labels into digital feature values; S3-2. Feature selection based on grey correlation analysis method: using the grey correlation analysis method to calculate the correlation between the basic electricity quantity and the feature data, and eliminating the feature values ​​with low correlation; S3-3. Predict basic electricity consumption based on multiple linear regression.

8. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 7 is characterized in that: The step S3-2 specifically includes: finding a difference sequence, finding a two-level difference, finding a correlation coefficient, and selecting relevant features according to the correlation degree: The difference sequence is the initialization of the reference sequence minus the corresponding number of each comparison sequence. The specific formula is: Δ i (k)=|X0(k)-X i (k)|; i = 1, 2, 3 ... n, In the formula, Δ i (k) is the absolute value of the difference between the comparison sequence and the reference sequence, X0(k) is the reference sequence, and X i (k) is a comparison sequence; The two-level difference is to find the maximum value range and the minimum value range. The specific formula is: Where M 极差 is the maximum range, m 极差 is the minimum range; The correlation coefficient is to find the average value of the correlation degree between the comparison sequence and the reference sequence at each point. The specific formula is: In the formula, ξ i (k) is the correlation value between the comparison sequence and the reference sequence at each point, Δ i (k) is the absolute value of the difference between the comparison sequence and the reference sequence, ρ is the resolution coefficient, which is 0.5, M is the maximum range, and m is the minimum range; The selecting of relevant features according to the correlation degree is: sorting the characteristic factors according to the size of the correlation coefficient, obtaining the correlation degree of the characteristic factors to the line loss, selecting relevant features, and eliminating irrelevant features.

9. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 7 is characterized in that: The step S3-3 is to substitute the characteristic value processed by the step S3-2 into the multivariate linear regression model to obtain the fitting result of the typical daily power and the characteristic value. The specific formula is: In the formula, the n-order polynomial has a0 to a n The n+1 unknown fitting coefficients, is the predicted value and x is the feature value.

10. The method for decomposing cooling and heating electricity based on kernel density estimation and multivariate linear regression according to claim 1 is characterized in that: The S5 step is to use the Pearson correlation coefficient method to calculate the correlation coefficient between the temperature and the cooling and heating electricity calculation results after the comparison benchmark value method, the basic electricity prediction method and the correction considering the power restriction measures. If the correlation coefficient after the correction considering the power restriction measures is higher than the correlation coefficient obtained by the comparison benchmark value method and the basic electricity prediction method, the step ends; if the correlation coefficient after the correction considering the power restriction measures is lower than the correlation coefficient obtained by the comparison benchmark value method and the basic electricity prediction method, then start again from the S1 step and reconstruct the feature label system.

Images