An AI-based energy consumption analysis system

Abstract

The present invention relates to the field of energy optimization technology, specifically to an AI-based energy consumption analysis system, the system comprising: a data acquisition and time-stamping module collects electricity, gas and water usage data from multiple monitoring points, adds a timestamp to each usage data, performs quality verification at the same time, eliminates data errors and eliminates duplicate records, and generates time-stamped energy data. In the present invention, the time window is adaptively adjusted to dynamically adjust according to actual data changes, so as to improve the response speed and accuracy of the system to energy usage fluctuations. By deeply analyzing abnormal consumption patterns, the solution can more accurately identify statistical deviations, and use these data to predict future energy needs, so as to make energy allocation more accurate and reduce waste. This comprehensive optimization process improves the automation and intelligence level of energy management, and helps to achieve the maximum efficiency and cost-effectiveness of energy use.

Description

Technical Field

[0001] The present invention relates to the technical field of energy optimization, and in particular to an energy consumption analysis system based on AI. Background Art

[0002] The energy optimization technology field focuses on improving the efficiency of energy use through advanced algorithms and control systems, aiming to reduce energy consumption and optimize the energy consumption process. This includes the use of automated systems, intelligent monitoring and predictive tools to analyze and manage energy flows. In commercial and industrial settings, energy optimization technology can help achieve cost savings, reduce carbon emissions and enhance system reliability. This technology field covers a wide range of applications such as smart grids, renewable energy integration, load management and building automation systems. As technology advances, these systems increasingly rely on artificial intelligence to predict energy demand and automatically adjust resource allocation to achieve optimal energy consumption efficiency.

[0003] Among them, AI-based energy consumption analysis systems refer to systems that use artificial intelligence technology to monitor, analyze and optimize energy use. The main purpose of such systems is to process and analyze large amounts of energy usage data through AI algorithms to identify energy-saving potential, predict future energy demand and automate energy management processes. In practical applications, such systems can help companies and organizations reduce energy consumption, reduce operating costs and support sustainable development goals. Through real-time data analysis and machine learning, these systems can provide detailed energy consumption reports and recommendations, making energy management more intelligent and efficient.

[0004] Although the automation systems in existing technologies can process energy data, they usually lack the ability to respond quickly and adapt to real-time changes. Fixed analysis models and time windows cannot effectively adapt to the rapid changes in energy use, which leads to insufficient timeliness and accuracy of data analysis. The lack of an effective abnormal monitoring mechanism also makes it difficult for traditional systems to identify and handle abnormal energy use in a timely manner, resulting in the inability to adjust energy allocation strategies in a timely manner, increasing unnecessary costs and waste of resources. These technical limitations show that although existing systems provide certain energy monitoring and management functions, their potential in improving energy efficiency and reducing environmental impact has not been fully realized. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an energy consumption analysis system based on AI.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: An AI-based energy consumption analysis system comprises:

[0007] The data collection and time stamping module collects electricity, gas and water usage data from multiple monitoring points, adds a time stamp to each usage data, and performs quality verification to eliminate data errors and duplicate records, and generates time-stamped energy data;

[0008] The time window adaptive adjustment module analyzes the change trend and frequency of the multi-energy data based on the time-tagged energy data, and automatically adjusts the time window to match the data fluctuation by real-time monitoring of the data change rate, thereby establishing an adaptive adjustment time window;

[0009] The abnormal pattern recognition module uses the data of the adaptively adjusted time window to perform abnormal point analysis, including detecting and identifying energy consumption data that statistically deviates from the normal pattern, classifying and recording the abnormal points, describing the abnormal trend of energy consumption, and obtaining abnormal consumption patterns;

[0010] The energy demand forecasting module uses the abnormal consumption pattern to perform demand forecasting, analyze the relationship between current abnormal data and demand forecasting methods, estimate energy demand changes in future time periods, and provide energy demand forecasting results.

[0011] As a further solution of the present invention, the step of acquiring the time-stamped energy data is specifically as follows:

[0012] Collect electricity, gas and water usage data from multiple monitoring points using the formula:

[0013]

[0014] Get the weighted original energy data set;

[0015] Where E represents the weighted original energy dataset, d i represents the data of the ith monitoring point, c i represents the data credibility coefficient of the i-th monitoring point, w i represents the weight factor, and p represents the adjustment index;

[0016] A timestamp is added to each piece of data in the weighted original energy data set, using the formula:

[0017]

[0018] Generate time-stamped energy data;

[0019] Among them, T d Represents the time-stamped energy data, e i represents a single data point in E, t i Represents data point e i The timestamp of , k represents the time adjustment coefficient;

[0020] The energy data marked with the timestamp is quality checked to eliminate data errors and duplicate records using the formula:

[0021]

[0022] When data i The value is 1 in case of error or duplication, otherwise it is 0, and the time-stamped energy data is obtained;

[0023] Among them, Q(T d ) represents time-stamped energy data, is a characteristic function of data errors or duplications, is a single data point marked with a timestamp, m i is the quality adjustment factor, and q is the strengthening factor.

[0024] As a further solution of the present invention, the steps for acquiring the change trend and frequency of the multi-energy data are specifically as follows:

[0025] From the time-stamped energy data, the difference between each pair of adjacent data points is calculated using the formula:

[0026]

[0027] The obtained difference array;

[0028] Among them, Q(T d ) represents the time-stamped energy data, ΔQ i Represents the data difference between the i-th time point and the i+1-th time point, Respectively represent adjacent time marker points;

[0029] Using the difference array ΔQ, calculate the absolute value of each difference, calculate the sum and average, using the formula:

[0030]

[0031] Get the average rate of change;

[0032] Where R represents the average rate of change and n represents the number of data points;

[0033] Analyze the change trend and frequency of multi-energy data by comparing each difference amount with the average change rate, using the formula:

[0034]

[0035] Get the changing trend and frequency of multi-energy data;

[0036] Among them, F represents the changing trend and frequency of multi-energy data, wi is the weight factor of the ith data point.

[0037] As a further solution of the present invention, the step of obtaining the adaptively adjusted time window is specifically:

[0038] Based on the changing trend and frequency of the multi-energy data, the volatility of the data is measured using the formula:

[0039]

[0040] Measure the statistical dispersion of the data and obtain volatility;

[0041] Among them, F i is the value of the i-th element in F, is the mean value of F, σ F is the standard deviation of F, V represents volatility, and n is the number of samples;

[0042] Using the volatility, the sensitive threshold of data fluctuation is calculated using the formula:

[0043]

[0044] Generate sensitivity thresholds;

[0045] Among them, k is the preset sensitivity coefficient, β is the adjustment index, and ∈ is a small amount to avoid division by zero;

[0046] Combined with the sensitivity threshold, the formula is adopted:

[0047]

[0048] Establish an adaptive adjustment time window T w ;

[0049] Here, δ is a small constant used for adjustment.

[0050] As a further solution of the present invention, the step of obtaining the abnormal consumption pattern is specifically as follows:

[0051] Using the data in the adaptively adjusted time window, the deviation of each data point relative to the mean value in the time window is calculated using the formula:

[0052]

[0053] Generate preliminary outlier labels;

[0054] Among them, x i represents the data point in the adaptive time window, μ represents the mean value of the data, σ represents the standard deviation of the data, T wrepresents the time window size obtained from the previous step, λ is the introduced smoothing parameter, and A i represents preliminary outlier labeling;

[0055] The outliers of the preliminary markers are screened using the formula:

[0056] E i ={x i ∣A i >τ·log(T w +γ)}

[0057] Filter the data points exceeding the threshold as significant outliers and generate a significant outlier set E i ;

[0058] Among them, τ is the pre-set threshold, γ is the logarithmic basis adjustment parameter, and E i Represents a set of significant outlier points;

[0059] The significant abnormal points are classified and recorded using the formula:

[0060]

[0061] Summarize the abnormal trends of energy consumption and obtain abnormal consumption patterns;

[0062] Where m is the number of outliers, w j The weight assigned based on the degree of abnormality, δ is the adjustment factor, α is the weighted index, and P represents the abnormal consumption pattern.

[0063] As a further solution of the present invention, the steps of obtaining the energy demand forecast result are specifically as follows:

[0064] The abnormal consumption pattern is used to analyze the abnormal current energy consumption using the formula:

[0065]

[0066] Generate trend analysis result D;

[0067] Among them, P i represents a data point in an abnormal consumption pattern, t i represents the weighting factor associated with time, σ P is the standard deviation of the data set, α and δ adjust the time weighting and data smoothness, and D represents the trend analysis result of the current data;

[0068] Based on the trend analysis results, a nonlinear regression model is used to estimate the change in energy demand in the future time period, using the formula:

[0069] F future=D·(1+β·log(Δt+∈))

[0070] Generate estimates of changes in energy demand;

[0071] Among them, β is the growth rate obtained based on historical data analysis, Δt is the length of the future time period, ∈∈ is a small positive number, which is used to make the logarithmic operation still valid when Δt=0, and F future To provide forecasts for changes in energy demand;

[0072] The results of the energy demand change estimates are summarized and formalized using the integral model formula:

[0073]

[0074] Integrate forecast data within the forecast period to provide energy demand forecast results;

[0075] Among them, F future (t) represents the predicted energy demand at time t, and the integral expresses the total demand over the entire forecast period, R forecast Represents the energy demand forecast result.

[0076] Compared with the prior art, the advantages and positive effects of the present invention are:

[0077] In the present invention, by collecting real-time data and marking timestamps for each record, the optimization scheme ensures that the data reflects the latest usage, and through continuous quality verification, effectively eliminates errors and duplications, and improves data quality. In addition, the adaptive adjustment time window is dynamically adjusted according to actual data changes, which improves the system's response speed and accuracy to fluctuations in energy use. Through in-depth analysis of abnormal consumption patterns, the scheme can more accurately identify statistical deviations and use these data to predict future energy needs, making energy allocation more accurate and reducing waste. This comprehensive optimization process improves the automation and intelligence level of energy management, and helps to maximize the efficiency and cost-effectiveness of energy use. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 is a system flow chart of the present invention;

[0079] Figure 2 A flow chart of the steps for obtaining time-stamped energy data of the present invention;

[0080] Figure 3 A flow chart of the steps for obtaining the changing trend and frequency of multi-energy data of the present invention;

[0081] Figure 4 A flow chart of the acquisition steps of adaptively adjusting the time window of the present invention;

[0082] Figure 5A flowchart of the steps for obtaining the abnormal consumption mode of the present invention;

[0083] Figure 6 This is a flow chart of the steps for obtaining the energy demand prediction results of the present invention. DETAILED DESCRIPTION

[0084] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0085] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0086] Embodiment 1

[0087] See also Figure 1 , an AI-based energy consumption analysis system includes:

[0088] The data collection and time stamping module collects electricity, gas and water usage data from multiple monitoring points, adds a time stamp to each usage data, and performs quality verification to eliminate data errors and duplicate records, and generates time-stamped energy data;

[0089] The time window adaptive adjustment module analyzes the changing trend and frequency of multi-energy data based on time-stamped energy data, and automatically adjusts the time window to match data fluctuations by real-time monitoring of data change rate, thus establishing an adaptive adjustment time window.

[0090] The abnormal pattern recognition module uses the data of the adaptively adjusted time window to perform abnormal point analysis, including detecting and identifying energy consumption data that statistically deviates from the normal pattern, classifying and recording the abnormal points, describing the abnormal trend of energy consumption, and obtaining abnormal consumption patterns;

[0091] The energy demand forecasting module uses abnormal consumption patterns to perform demand forecasting, analyzes the relationship between current abnormal data and demand forecasting methods, estimates changes in energy demand in future time periods, and provides energy demand forecasting results.

[0092] Time-stamped energy data includes electricity data timestamp, gas data timestamp, and water resource data timestamp. Adaptive adjustment of time window specifically refers to time window size adjustment, data fluctuation response, and time window efficiency. Abnormal consumption pattern includes abnormal data type, consumption deviation level, and abnormal pattern frequency. Energy demand forecasting results specifically include predicted demand change, demand change trend, and forecast accuracy.

[0093] See also Figure 2 ,The steps for acquiring time-stamped energy data are as follows:

[0094] Collect electricity, gas and water usage data from multiple monitoring points using the formula:

[0095]

[0096] Get the weighted original energy data set;

[0097] Where E represents the weighted original energy dataset, d i represents the data of the ith monitoring point, c i represents the data credibility coefficient of the i-th monitoring point, w i represents the weight factor, and p represents the adjustment index;

[0098] Add a timestamp to each piece of data in the weighted raw energy dataset using the formula:

[0099]

[0100] Generate time-stamped energy data;

[0101] Among them, T d Represents the time-stamped energy data, e i represents a single data point in E, t i Represents data point e i The timestamp of , k represents the time adjustment coefficient;

[0102] The energy data marked with timestamp is quality checked to eliminate data errors and duplicate records using the formula:

[0103]

[0104] When data i The value is 1 in case of error or duplication, otherwise it is 0, and the time-stamped energy data is obtained;

[0105] Among them, Q(T d ) represents time-stamped energy data, is a characteristic function of data errors or duplications, is a single data point marked with a timestamp, mi is the quality adjustment factor, and q is the strengthening factor.

[0106] Step 1: Collect data and calculate weighted raw energy dataset

[0107] formula:

[0108]

[0109] Parameter details:

[0110] d i : Data of the ith monitoring point (e.g., power consumption value)

[0111] c i : The data credibility coefficient of the i-th monitoring point, reflecting the data quality (assumed to be between 0.8 and 1.0)

[0112] w i : The weight factor of the ith monitoring point, assigned according to the importance of the monitoring point (assumed to be between 1 and 5)

[0113] p: adjustment index, used to adjust the influence of the weight factor (assuming the value is 2)

[0114] Calculation process:

[0115] For each monitoring point i, obtain its energy consumption d i , data credibility c i , and weight factor w i .

[0116] Calculate the contribution of each monitoring point:

[0117] The contributions of all monitoring points are aggregated to obtain the total energy data E.

[0118] Calculation example:

[0119] Assume there are 3 monitoring points, and the data of each point are as follows:

[0120] Monitoring point 1: d1 = 100 kWh, c1 = 0.95, w1 = 2

[0121] Monitoring point 2: d2 = 150 kWh, c2 = 0.90, w2 = 3

[0122] Monitoring point 3: d3 = 200 kWh, c3 = 0.98, w3 = 4

[0123] calculate:

[0124] Contribution of monitoring point 1: 100×0.95×2 2 =380

[0125] Monitoring point 2 contribution: 150×0.90×3 2 =1215

[0126] Monitoring point 3 contribution: 200×0.98×4 2 =3136

[0127] Total E = 380 + 1215 + 3136 = 4731 kWh

[0128] Step 2: Timestamp

[0129] formula:

[0130]

[0131] Parameter details:

[0132] e i : The i-th data point, the energy data obtained from step 1

[0133] t i : Data point e i Timestamp

[0134] k: time adjustment coefficient (assumed value is 0.5)

[0135] Calculation process:

[0136] For each data point e i , get its timestamp t i .

[0137] Calculate the time adjustment value for each data point: Aggregate the time-adjusted values ​​of all data points to generate the timestamped energy data T d . Calculation example:

[0138] Continuing with the above data, assume that the timestamp is t i 10, 20, 30 minutes respectively:

[0139] t1=10, t2=20, t3=30

[0140] T d calculate:

[0141] Data point 1 time adjustment value: 380×10 (1 / 0.5) =380×100=38000 Data point 2 time adjustment value: 1215×20 (1 / 0.5) =1215×400=486000 Data point 3 time adjustment value: 3136×30 (1 / 0.5) =3136×900=2822400Total T d=38000+486000+2822400=3312400 time adjustment kWh Step 3: Quality verification and final result generation

[0142] formula:

[0143]

[0144] Parameter details:

[0145] Indicative function of data error or duplication, when e i 0 if true, 1 otherwise

[0146] The time adjustment value for each data point in step 2

[0147] m i : Quality adjustment factor, reflecting the stability of data quality (assuming a value of 10) q: Strengthening factor, increasing the strictness of data verification (assuming a value of 1.5)

[0148] Calculation process:

[0149] For each data point Marked by its quality Perform quality inspection.

[0150] Calculate the quality check value for each data point: Sum up the quality check values ​​of all data points to get the final time-stamped energy data D f . Calculation example:

[0151] Assume that all data points are correct, i.e. D f calculate:

[0152] Data point 1 quality check value:

[0153] Data point 2 quality check value:

[0154] Data point 3 quality check value:

[0155] Total D f =22949.1+1586225.2+12915644.3=14532818.6 final verification value kWh.

[0156] See also Figure 3 ,The specific steps for obtaining the changing trend and frequency of multi-energy data are:

[0157] From the time-stamped energy data, calculate the difference between each pair of adjacent data points using the formula:

[0158]

[0159] The obtained difference array;

[0160] Among them, Q(T d ) represents the time-stamped energy data, ΔQ i Represents the data difference between the i-th time point and the i+1-th time point, Respectively represent adjacent time marker points;

[0161] Using the difference array ΔQ, calculate the absolute value of each difference, calculate the sum and average, using the formula:

[0162]

[0163] Get the average rate of change;

[0164] Where R represents the average rate of change and n represents the number of data points;

[0165] Analyze the change trend and frequency of multi-energy data by comparing each difference amount with the average change rate, using the formula:

[0166]

[0167] Get the changing trend and frequency of multi-energy data;

[0168] Among them, F represents the changing trend and frequency of multi-energy data, w i is the weight factor of the ith data point.

[0169] Formula for step 1:

[0170]

[0171] Parameter details:

[0172] and Represents the time points and energy data.

[0173] ΔQ i Represents the difference in data between these two time points.

[0174] Derivation and examples:

[0175] Assume that at time The energy data is 150 units, and in time The data is 160 units, and the calculation process is as follows:

[0176] ΔQ1=160-150=10

[0177] This result, 10 units, represents the arrive The amount of energy data added during a time interval.

[0178] Formula for step 2:

[0179]

[0180] Parameter details:

[0181] |ΔQ i | is the absolute value of the change in the ith data point.

[0182] n is the total number of data points.

[0183] R is the calculated average rate of change.

[0184] Derivation and examples:

[0185] Continuing with the above assumptions, assume there are three changes: 10, -5, 15. The total number n = 3, and the calculation process is as follows:

[0186]

[0187] This result, 10 units, represents the average change in energy data per time interval.

[0188] Formula for step 3:

[0189]

[0190] Parameter details:

[0191] w i is the weight factor of the ith data point.

[0192] |ΔQ i The definitions of | and R are the same as before.

[0193] F represents the weighted changing trend and frequency of multi-energy data.

[0194] Derivation and examples:

[0195] Using the above changes and assuming weight factors w1 = 0.5, w2 = 1.0, w3 = 1.5, the calculation process is as follows:

[0196]

[0197] This result, 3.25, represents the weighted trend and frequency of change, with larger values ​​indicating more significant changes.

[0198] See also Figure 4 , the specific steps for obtaining the adaptive adjustment time window are:

[0199] Based on the changing trend and frequency of multi-energy data, the volatility of data is measured using the formula:

[0200]

[0201] Measure the statistical dispersion of the data and obtain volatility;

[0202] Among them, F i is the value of the i-th element in F, is the mean value of F, σ F is the standard deviation of F, V represents volatility, and n is the number of samples;

[0203] Using volatility, calculate the sensitive threshold of data fluctuation using the formula:

[0204]

[0205] Generate sensitivity thresholds;

[0206] Among them, k is the preset sensitivity coefficient, β is the adjustment index, and ∈ is a small amount to avoid division by zero;

[0207] Combined with the sensitive threshold, the formula is used:

[0208]

[0209] Establish an adaptive adjustment time window T w ;

[0210] Here, δ is a small constant used for adjustment.

[0211] Step 1: Volatility Calculation

[0212] formula:

[0213]

[0214] Parameter explanation:

[0215] F: Dataset of changing trends and frequencies

[0216] F i : The value of the i-th element in F The average value of F

[0217] σ F : Standard deviation of F

[0218] V: Volatility

[0219] n: number of data points

[0220] Calculation process and examples:

[0221] Assume there is an F array with 5 data points: F = [5,7,6,8,10] calculation

[0222]

[0223] Calculate the standard deviation σ F :

[0224]

[0225] Calculate volatility V:

[0226]

[0227] Step 2: Calculation of sensitivity threshold

[0228] formula:

[0229]

[0230] Parameter explanation:

[0231] k: sensitivity coefficient preset by system requirements β: adjustment index

[0232] ∈: small constant to avoid division by zero V: volatility

[0233] T threshold : Sensitive threshold

[0234] Calculation process and examples:

[0235] Assume k = 1.5, β = 2, and ∈ = 0.01:

[0236]

[0237] Step 3: Adaptively adjust the calculation formula of the time window:

[0238]

[0239] Parameter explanation:

[0240] T threshold： : Sensitive threshold δ: A small constant used for adjustment

[0241] T w : Time window

[0242] Calculation process and examples:

[0243] Assume δ = 0.005:

[0244]

[0245] Result description:

[0246] The volatility V = 0.9993 indicates the degree of variation in the data set.

[0247] Sensitivity threshold T threshold =1.4868 represents the system's sensitivity to fluctuations.

[0248] Time window T w The logarithmic value of = -0.3997 represents the time window size adjusted based on the sensitivity threshold, which is used for real-time monitoring and processing of data.

[0249] See also Figure 5 , the specific steps for obtaining the abnormal consumption mode are:

[0250] Using the data in the adaptively adjusted time window, calculate the deviation of each data point relative to the mean value within the time window using the formula:

[0251]

[0252] Generate preliminary outlier labels;

[0253] Among them, x i represents the data point in the adaptive time window, μ represents the mean value of the data, σ represents the standard deviation of the data, T w represents the time window size obtained from the previous step, λ is the introduced smoothing parameter, and A i represents preliminary outlier labeling;

[0254] The initially marked outliers are screened using the formula:

[0255] E i ={x i ∣A i >τ·log(T w +γ)}

[0256] Filter the data points exceeding the threshold as significant outliers and generate a significant outlier set E i ;

[0257] Among them, τ is the pre-set threshold, γ is the logarithmic basis adjustment parameter, and E i Represents a set of significant outlier points;

[0258] Classify and record the significant outliers using the formula:

[0259]

[0260] Summarize the abnormal trends of energy consumption and obtain abnormal consumption patterns;

[0261] Where m is the number of outliers, w j The weight assigned based on the degree of abnormality, δ is the adjustment factor, α is the weighted index, and P represents the abnormal consumption pattern.

[0262] Step 1: Compute preliminary outlier labels

[0263] formula:

[0264]

[0265] Parameter explanation and calculation process:

[0266] x i : The i-th data point in the time window T w The value in .

[0267] μ: data point x i The average value of .

[0268] σ: data point x i The standard deviation of .

[0269] T w : The size of the time window.

[0270] λ: Smoothing parameter used to adjust the sensitivity of the deviation.

[0271] Derivation process:

[0272] Suppose there are 5 data points in a given time window: 10, 12, 9, 11, 13. Based on this data, we can calculate:

[0273]

[0274] The calculation of σ involves taking the square root of the variance, which is the average of the sum of the squares of the differences between most values ​​and the mean, that is:

[0275]

[0276] T w Assume it is 5 (the time window contains 5 data points).

[0277] λ is set to 0.5 to increase calculation stability.

[0278] For x i =10 outlier markers:

[0279]

[0280] This calculation shows how to calculate preliminary outlier labels for each data point based on actual data.

[0281] Step 2: Screening for significant outliers

[0282] formula:

[0283] E i ={x i ∣A i >τ·log(T w +γ)}

[0284] Parameter explanation and calculation process:

[0285] τ: threshold, assumed to be 0.2.

[0286] γ: Logarithmic basis adjustment parameter, set to 1 to simplify calculations.

[0287] Derivation process:

[0288] According to the previous T w and γ,

[0289] log(T w +γ)=log(5+1)=log(6)≈0.778

[0290] τ·log(T w +γ)=0.2·0.778=0.156

[0291] From step 1, we know that A i =0.201 is greater than 0.156, so x i =10 was considered a significant abnormality.

[0292] Step 3: Identify abnormal consumption patterns

[0293] formula:

[0294]

[0295] Parameter explanation and calculation process:

[0296] m: The number of outliers, here it is 1.

[0297] w j : The weight of the outlier point, assuming that the weight of each point is 1.

[0298] δ: adjustment factor, set to 0.1.

[0299] α: weighting exponent, set to 2.

[0300] Derivation process:

[0301] Assumption E j=10,

[0302]

[0303] This result P = 1.95 represents the abnormal consumption pattern index obtained after integrating all significant abnormal data points, indicating the degree of influence of these abnormal points on energy consumption after referring to the time window and the weight of the abnormal points.

[0304] See also Figure 6 , the specific steps for obtaining energy demand forecast results are:

[0305] The abnormal consumption mode is used to analyze the current energy consumption abnormality using the formula:

[0306]

[0307] Generate trend analysis result D;

[0308] Among them, P i represents a data point in an abnormal consumption pattern, t i represents the weighting factor associated with time, σ P is the standard deviation of the data set, α and δ adjust the time weighting and data smoothness, and D represents the trend analysis result of the current data;

[0309] Based on the trend analysis results, a nonlinear regression model is used to estimate the change in energy demand in the future time period, using the formula:

[0310] F future =D·(1+β·log(Δt+∈))

[0311] Generate estimates of changes in energy demand;

[0312] Among them, β is the growth rate obtained by analyzing historical data, Δt is the length of the future time period, ∈ is a small positive number, and the logarithmic operation is still valid when Δt = 0. future To provide forecasts for changes in energy demand;

[0313] The results of the energy demand change estimates are summarized and formalized using the integral model formula:

[0314]

[0315] Integrate forecast data within the forecast period to provide energy demand forecast results;

[0316] Among them, F future (t) represents the predicted energy demand at time t, and the integral expresses the total demand over the entire forecast period, R forecast Represents the energy demand forecast result.

[0317] Formula and derivation of step 1

[0318] formula:

[0319]

[0320] Parameter explanation:

[0321] P i : The energy consumption value of the ith data point in the abnormal consumption pattern.

[0322] t i : The time factor corresponding to the i-th data point.

[0323] σ P : All P i The standard deviation of the values.

[0324] α: time weighting factor.

[0325] δ: Smoothing parameter, to avoid the denominator being zero.

[0326] Calculation process and examples:

[0327] Assume there are 5 data points, the abnormal consumption patterns P are 10, 20, 15, 25, 5 units of energy, the time factors t are 1, 2, 3, 4, 5 hours, and α = 2 is selected as the time weighting factor. The calculated σ P ≈7.07 (the definition formula of standard deviation is used in actual calculation), and the smoothing parameter δ is set to 1.

[0328] Compute the contribution of the first data point:

[0329]

[0330] Similarly, calculate the other points and sum them:

[0331]

[0332] This value D represents the abnormal energy consumption trend index after time weighting and smoothing.

[0333] Formula and derivation of step 2

[0334] formula:

[0335] F future =D·(1+β·log(Δt+∈))

[0336] Parameter explanation:

[0337] D: Trend analysis results obtained from step 1.

[0338] β: predicted growth rate, based on historical data analysis.

[0339] Δt: Length of the prediction period.

[0340] ∈: positive number to avoid errors in logarithm calculation.

[0341] Calculation process and examples:

[0342] Assume that the energy demand for the next 10 years is predicted, and set β = 0.05, ∈ = 0.01.

[0343] F future =232.59·(1+0.05·log(10+0.01))

[0344] log(10.01)≈1

[0345] F future =232.59·(1+0.05·1)=232.59·1.05≈244.22

[0346] This value F future Represents the time-adjusted forecast of future energy demand.

[0347] Formula and derivation of step 3

[0348] formula:

[0349]

[0350] Parameter explanation:

[0351] F future (t): predicted energy demand at time t.

[0352] Calculation process and examples:

[0353]

[0354] This value R forecast Represents the total energy demand forecast for the next 10 years.

[0355] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An AI-based energy consumption analysis system, characterized in that: The system comprises: The data collection and time stamping module collects electricity, gas and water usage data from multiple monitoring points, adds a time stamp to each usage data, and performs quality verification to eliminate data errors and duplicate records, and generates time-stamped energy data; The time window adaptive adjustment module analyzes the change trend and frequency of the multi-energy data based on the time-tagged energy data, and automatically adjusts the time window to match the data fluctuation by real-time monitoring of the data change rate, thereby establishing an adaptive adjustment time window; The abnormal pattern recognition module uses the data of the adaptively adjusted time window to perform abnormal point analysis, including detecting and identifying energy consumption data that statistically deviates from the normal pattern, classifying and recording the abnormal points, describing the abnormal trend of energy consumption, and obtaining abnormal consumption patterns; The energy demand forecasting module uses the abnormal consumption pattern to perform demand forecasting, analyze the relationship between the current abnormal data and the demand forecasting method, estimate the energy demand changes in the future time period, and provide energy demand forecasting results; The steps for obtaining the abnormal consumption pattern are specifically as follows: Using the data in the adaptively adjusted time window, the deviation of each data point relative to the mean value in the time window is calculated using the formula: Generate preliminary outlier labels; Among them, x i represents the data point in the adaptive time window, μ represents the mean value of the data, σ represents the standard deviation of the data, T w represents the time window size obtained from the previous step, λ is the introduced smoothing parameter, and A i represents preliminary outlier labeling; The outliers of the preliminary markers are screened using the formula: E i ={x i ∣A i >τ·log(T w +c)} Filter the data points exceeding the threshold as significant outliers and generate a significant outlier set E i ; Among them, τ is the pre-set threshold, γ is the logarithmic basis adjustment parameter, and E i Represents a set of significant outlier points; The significant abnormal points are classified and recorded using the formula: Summarize the abnormal trends of energy consumption and obtain abnormal consumption patterns; Where m is the number of outliers, w j The weights assigned based on the degree of abnormality, δ is the adjustment factor, α is the weighted index, and P represents the abnormal consumption pattern; The steps for obtaining the energy demand forecast result are specifically as follows: The abnormal consumption pattern is used to analyze the abnormal current energy consumption using the formula: Generate trend analysis result D; Among them, P i represents a data point in an abnormal consumption pattern, t i represents the weighting factor associated with time, σ P is the standard deviation of the data set, α and δ adjust the time weighting and data smoothness, and D represents the trend analysis result of the current data; Based on the trend analysis results, a nonlinear regression model is used to estimate the change in energy demand in the future time period, using the formula: F future =D·(1+β·log(Δt+∈)) Generate estimates of changes in energy demand; Among them, β is the growth rate obtained by analyzing historical data, Δt is the length of the future time period, ∈ is a small positive number, and the logarithmic operation is still valid when Δt = 0. future To provide forecasts for changes in energy demand; The results of the energy demand change estimates are summarized and formalized using the integral model formula: Integrate forecast data within the forecast period to provide energy demand forecast results; Among them, F future (t) represents the predicted energy demand at time t, and the integral expresses the total demand over the entire forecast period, R forecast Represents the energy demand forecast result.

2. The AI-based energy consumption analysis system according to claim 1, characterized in that: The steps for obtaining the time-stamped energy data are specifically as follows: Collect electricity, gas and water usage data from multiple monitoring points using the formula: Get the weighted original energy data set; Where E represents the weighted original energy dataset, d i represents the data of the ith monitoring point, c i represents the data credibility coefficient of the i-th monitoring point, w i represents the weight factor, and p represents the adjustment index; A timestamp is added to each piece of data in the weighted original energy data set, using the formula: Generate time-stamped energy data; Among them, T d Represents the time-stamped energy data, e i represents a single data point in E, t i Represents data point e i The timestamp of , k represents the time adjustment coefficient; The energy data marked with the timestamp is quality checked to eliminate data errors and duplicate records using the formula: When data i The value is 1 in case of error or duplication, otherwise it is 0, and the time-stamped energy data is obtained; Among them, Q(T d ) represents time-stamped energy data, is a characteristic function of data errors or duplications, is a single data point marked by a timestamp, m i is the quality adjustment factor, and q is the strengthening factor.

3. The AI-based energy consumption analysis system according to claim 2, characterized in that: The steps for obtaining the change trend and frequency of the multi-energy data are specifically as follows: From the time-stamped energy data, the difference between each pair of adjacent data points is calculated using the formula: The obtained difference array; Among them, Q(T d ) represents the time-stamped energy data, ΔQ i Represents the data difference between the i-th time point and the i+1-th time point, Represent adjacent time marker points respectively; Using the difference array ΔQ, calculate the absolute value of each difference, calculate the sum and average, using the formula: Get the average rate of change; Where R represents the average rate of change and n represents the number of data points; Analyze the change trend and frequency of multi-energy data by comparing each difference amount with the average change rate, using the formula: Get the changing trend and frequency of multi-energy data; Among them, F represents the changing trend and frequency of multi-energy data, w i is the weight factor of the ith data point.

4. The AI-based energy consumption analysis system according to claim 3, characterized in that: The steps of obtaining the adaptive adjustment time window are specifically as follows: Based on the changing trend and frequency of the multi-energy data, the volatility of the data is measured using the formula: Measure the statistical dispersion of the data and obtain volatility; Among them, F i is the value of the i-th element in F, is the mean value of F, σ F is the standard deviation of F, V represents volatility, and n is the number of samples; Using the volatility, the sensitive threshold of data fluctuation is calculated using the formula: Generate sensitivity thresholds; Among them, k is the preset sensitivity coefficient, β is the adjustment index, and ∈ is a small amount to avoid division by zero; Combined with the sensitivity threshold, the formula is adopted: Establish an adaptive adjustment time window T w ; Here, δ is a small constant used for adjustment.

Images