Artificial intelligence-based power demand data analysis method and system

Abstract

The application discloses a power demand data analysis method and system based on artificial intelligence, and belongs to the technical field of smart grid and power system automation, and comprises the following steps: collecting multi-source heterogeneous data and preprocessing the data; using a causal discovery algorithm to analyze the causal relationship between the multi-source data, eliminating irrelevant factors to select final variables, and outputting a causal reasoning result; constructing a deep learning model and outputting a prediction result; and optimizing a power dispatching strategy according to the prediction result. Through time alignment and abnormal detection mechanisms of multi-source heterogeneous data, the application effectively solves the time sequence misalignment problem in multi-protocol data fusion, and improves data quality; in combination with a causal discovery algorithm and statistical testing, the application accurately identifies real causal relationships and eliminates false correlation interference in traditional correlation analysis; and in combination with a time sequence neural network and a causal reasoning model, the application enhances the interpretability of the model while retaining the time sequence characteristics of historical data, and significantly improves the prediction robustness under extreme working conditions.

Description

Technical Field

[0001] This invention relates to the field of smart grid and power system automation technology, specifically to a method and system for analyzing power demand data based on artificial intelligence. Background Technology

[0002] Against the backdrop of the rapid development of smart grids, the accuracy of electricity demand forecasting directly impacts grid dispatch efficiency and operational economy. Traditional methods typically rely on statistical analysis of historical load data, combined with environmental factors such as temperature and seasons to construct forecasting models (e.g., ARIMA, Support Vector Machines). However, with the explosive growth of multi-source heterogeneous data, existing technologies face significant bottlenecks: First, in terms of data integration, multi-protocol data generated by smart meters, SCADA systems, and IoT sensors exhibit differences in temporal granularity. Existing time alignment methods often employ fixed-frequency resampling, leading to the loss of high-frequency data details, especially prone to synchronization errors during periods of sudden load changes. Second, in terms of feature selection, traditional Pearson correlation coefficients for variable selection are susceptible to confounding factors (e.g., spurious correlations between economic indicators and electricity consumption). While mainstream LSTM forecasting models can capture time-series features, they lack causal inference mechanisms, resulting in forecasting errors fluctuating by more than 25% when encountering extreme weather or sudden power outages. Finally, in the dispatch optimization stage, existing gradient descent methods often use fixed penalty coefficients, making it difficult to dynamically adapt to load fluctuations, leading to an increase in supply-demand imbalance rates during peak hours. Therefore, there is an urgent need for an intelligent analysis system that integrates causal discovery and dynamic optimization to solve systemic problems such as multi-source data fusion, spurious correlation of variables, insufficient model generalization ability, and rigid scheduling strategies. Summary of the Invention

[0003] To address the aforementioned technical challenges, an artificial intelligence-based method for analyzing electricity demand data is proposed, which includes collecting multi-source heterogeneous data and preprocessing the data.

[0004] Based on the preprocessed data, a causal discovery algorithm is used to analyze the causal relationships between multi-source data, remove irrelevant factors, select the final variables, and output the causal inference results.

[0005] Build a deep learning model, optimize it based on causal reasoning results, and output prediction results;

[0006] Optimize power dispatching strategies based on forecast results;

[0007] The method of using causal discovery algorithms to analyze causal relationships among multi-source data includes performing variable independence tests using the PC algorithm, removing redundant variables when the conditional independence value is below a threshold, inferring causal directions using a linear non-Gaussian model, and optimizing the causal structure using a greedy search algorithm.

[0008] The optimized power dispatch strategy includes monitoring the continuous deviation between actual power supply and predicted values, triggering multi-source data recalibration and causal graph updates, using LSTM to verify abnormal data and generating corrected predictions, optimizing through gradient descent and issuing new dispatch instructions.

[0009] As a preferred embodiment of the artificial intelligence-based power demand data analysis method of the present invention, the step of collecting multi-source heterogeneous data and preprocessing the data includes:

[0010] Load data is collected through industrial communication protocols, received in real time using a distributed stream processing platform, stored in a time-series database, and the electricity consumption growth rate of industrial and commercial users is obtained through power company reports.

[0011] Power grid status data is obtained from the SCADA system via the IEC61850 communication protocol, and power monitoring is performed using a synchronous phasor measurement device.

[0012] The preprocessing includes synchronizing the time of multi-source data to a unified time step using time alignment technology, synchronizing the time of low-frequency data using linear interpolation to ensure consistent time steps across data sources, processing outliers based on the statistical 3σ criterion, and normalizing each data point using Min-Max.

[0013] As a preferred embodiment of the artificial intelligence-based electricity demand data analysis method of the present invention, the step of analyzing the causal relationships between multi-source data using a causal discovery algorithm based on preprocessed data includes:

[0014] Based on the preprocessed multi-source heterogeneous data, candidate variables related to electricity demand are initially selected through physical mechanisms. The independence of causal relationships is then tested using the PC algorithm to eliminate redundant variables.

[0015] The candidate variables include environmental variables and power grid operation variables;

[0016] Environmental variables, including power grid operation variables, include temperature, humidity, power grid load, and power supply status;

[0017] When the value of the conditional independence test is less than the set threshold, the current events are independent under the given conditions. The direction of causal relationship is inferred using a linear non-Gaussian model, and optimization is carried out in combination with causal intervention strategies.

[0018] The optimal causal structure is searched among all causal relationships using a greedy equivalence search, outputting a maximally optimized causal graph. The formula for optimizing the causal structure is expressed as:

[0019]

[0020] Among them, P(D)i |G) represents data D i The probability D under the causal graph G i Let represent the i-th data point, G represent the causal graph, and argmax represent the argmax value. G The expression represents selecting the G among all causal graphs G that maximizes the subsequent expression.

[0021] As a preferred embodiment of the artificial intelligence-based electricity demand data analysis method described in this invention, the step of eliminating irrelevant factors and selecting the final variable includes:

[0022] Use defined intervention operations to remove spurious correlation factors, determine the expected value after intervention through intervention operations, calculate the influence between causality, output the causal influence with low correlation, remove variables with low correlation but no causal relationship through Pearson correlation coefficient, and set a correlation threshold to remove data with correlation below the threshold.

[0023] The variables with causal relationships, after eliminating spurious correlations, are scored. The importance scoring formula is expressed as follows:

[0024] S(X)=w1·|ρ(X,Y)|+w2·I(Y|do(X))

[0025] Where S(X) represents the importance score of variable X, w1 and w2 represent weight parameters, and I(Y|do(X)) represents the amount of information about how variable X affects Y through causal intervention. The top k variables with the highest S(X) are selected to form the variable dataset X. * X * ={X1,X2,...,X k}

[0026] As a preferred embodiment of the artificial intelligence-based electricity demand data analysis method of the present invention, wherein: the construction of a deep learning model, optimization based on causal reasoning results, and output of prediction results include,

[0027] A new coefficient is introduced to adjust the tolerance for scheduling errors. A deep learning model is constructed, which optimizes power scheduling based on power demand values, minimizes power supply costs, and reduces power demand prediction errors. By combining power supply costs with scheduling error penalty and scheduling flexibility control, power demand scheduling is optimized.

[0028] A deep learning model is constructed using a temporal neural network model. The causal inference results obtained are combined with the deep learning model to capture the time series patterns of electricity demand and enhance the predictive ability of the deep learning model by utilizing causal relationships.

[0029] As a preferred embodiment of the artificial intelligence-based power demand data analysis method of the present invention, the step of optimizing the power dispatch strategy based on the prediction results includes:

[0030] The power dispatch optimization problem is solved by combining nonlinear programming with gradient descent. The constraints are set as capacity limits of power supply and balance constraints of system load. Power dispatch is implemented according to the optimized power supply plan so that the difference between actual power supply and predicted power demand approaches 0.

[0031] After the dispatch is implemented, the actual operation of the power system is continuously monitored. If the error exceeds the set threshold, the dispatch strategy is adjusted based on real-time feedback information.

[0032] As a preferred embodiment of the artificial intelligence-based power demand data analysis method of the present invention, the step of adjusting the scheduling strategy based on real-time feedback information includes:

[0033] The system monitors the deviation between the actual power supply and the predicted demand in real time. When the percentage deviation exceeds the threshold for the corresponding area type for a number of consecutive preset time periods, a dynamic adjustment mechanism is triggered.

[0034] The current real-time data is re-verified, and the time synchronization calibration of multi-source data is performed using linear interpolation. The abnormal power supply data is reconstructed and verified using a long short-term memory network.

[0035] The causal relationship graph is updated based on the re-verified data. New causal variables are selected through causal discovery algorithms. The updated causal variables are input into the demand forecasting model to generate revised electricity demand forecasts. The gradient descent method is used to optimize the scheduling of the revised forecasts, generating new power supply schemes. The optimized supply schemes are then distributed to distribution automation equipment for execution through the power system communication protocol.

[0036] Another objective of this invention is to provide an artificial intelligence-based power demand data analysis system that solves the problems of low data integration efficiency, inaccurate causal inference, poor model generalization ability, and delayed dispatch response in existing power forecasting through a multi-module collaborative intelligent analysis architecture.

[0037] To address the aforementioned technical problems, this invention provides the following technical solution: an artificial intelligence-based power demand data analysis system, comprising: a multi-source data governance module, a causal inference engine module, a time-series-causal fusion prediction module, and an adaptive scheduling optimization module;

[0038] The multi-source data governance module uses a time alignment engine and an anomaly detection unit to achieve cross-source data synchronization and quality improvement.

[0039] The causal inference engine module is based on a hybrid architecture of PC algorithm and linear non-Gaussian model. It performs variable independence test and causal direction inference, and combines greedy search to generate the optimal causal graph to eliminate spurious correlation interference.

[0040] The time-series-causal fusion prediction module extracts historical load time-series features through an LSTM network and dynamically weights and fuses them with the key variable weights output by causal inference to form an interpretable composite prediction model.

[0041] The adaptive scheduling optimization module integrates a nonlinear programming solver and a real-time feedback mechanism. It dynamically adjusts the power supply strategy based on the prediction results, iteratively optimizes the supply and demand matching through the gradient descent algorithm, and triggers the system self-repair process based on deviation monitoring.

[0042] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the artificial intelligence-based power demand data analysis method.

[0043] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the artificial intelligence-based power demand data analysis method.

[0044] The beneficial effects of this invention are as follows: This invention effectively solves the problem of temporal misalignment during multi-protocol data fusion by using a time alignment and anomaly detection mechanism for multi-source heterogeneous data, thereby improving data quality; it accurately identifies true causal relationships by combining causal discovery algorithms and statistical tests, eliminating spurious correlation interference in traditional correlation analysis; it integrates temporal neural networks and causal inference models, enhancing the interpretability of the model while preserving the temporal characteristics of historical data, and significantly improving the prediction robustness under extreme operating conditions; it introduces a dynamic scheduling optimization strategy, which adaptively adjusts the prediction model and scheduling parameters through a real-time feedback mechanism, breaking through the limitations of traditional fixed threshold control and realizing intelligent closed-loop control of power supply and demand balance. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 The above is a flowchart of an artificial intelligence-based power demand data analysis method provided in one embodiment of the present invention. Detailed Implementation

[0047] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0048] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides an artificial intelligence-based method for analyzing electricity demand data, including:

[0049] Collect heterogeneous data from multiple sources and preprocess the data.

[0050] Based on the preprocessed data, a causal discovery algorithm is used to analyze the causal relationships between multiple data sources, eliminate irrelevant factors, select the final variables, and output the causal inference results.

[0051] Build a deep learning model, optimize it based on the causal reasoning results, and output the prediction results.

[0052] Optimize power dispatching strategies based on forecast results.

[0053] Load data is read from smart meters, substation monitoring equipment, and distribution automation systems using the MODBUS / IEC104 protocol. The data is received in real time using the Kafka streaming framework and stored in a time-series database.

[0054] Deploy IoT sensors in industrial and residential areas to collect meteorological data in real time.

[0055] Obtain the electricity consumption growth rate of industrial and commercial users through power company reports.

[0056] Power grid status data is acquired from the SCADA system via the IEC 61850 communication protocol, and high-precision power monitoring is performed using a synchronous phasor measurement device. Due to the heterogeneity of the collected data, standardization is required. Time alignment technology is used to synchronize the time of multi-source data to a unified time step, ensuring data consistency and avoiding data errors caused by time mismatch.

[0057] In scenarios involving multi-source data acquisition, different data sources have different time granularities. Smart meter data is collected at a higher frequency, while meteorological data is collected at a lower frequency, resulting in different data time sequences. To ensure time consistency among different data sources, linear interpolation is used to synchronize the time of low-frequency data, ensuring that the time steps of each data source are consistent.

[0058] This approach ensures the temporal consistency of multi-source data used in subsequent analyses, avoids the impact of time misalignment on analysis results, and improves data quality and analytical accuracy.

[0059] Outlier handling uses the statistical 3σ criterion to detect extreme values ​​in the data, effectively removing erroneous data and improving data quality.

[0060] The mean and standard deviation of the dataset were calculated, and the researchers set a threshold of mean ± 3 times the standard deviation; data exceeding this range were considered outliers. An LSTM autoencoder was used to model the outlier-removed data to further validate its validity.

[0061] Outlier handling not only removes outlier data points, but also allows for further modeling and verification of the data's validity through LSTM autoencoders, improving data accuracy and providing more reliable input for subsequent analysis.

[0062] Normalization transforms data into the same dimensional range, making the influence of all features on the model relatively balanced.

[0063] For each data point, the data point is normalized using Min-Max.

[0064] Normalized data has a unified range, which helps improve the training efficiency and accuracy of machine learning models and avoids bias caused by inconsistent feature dimensions. In particular, when using deep learning models, normalization helps to speed up convergence and improve stability.

[0065] Based on the collected multi-source heterogeneous data, a causal relationship graph is constructed for causal inference. First, candidate variables related to electricity demand are initially selected using expert knowledge and physical mechanisms. Then, the independence of causal relationships is tested using a probabilistic computation (PC) algorithm to eliminate redundant variables and obtain a more accurate causal relationship structure.

[0066] The relevant candidate variables include environmental factors, socioeconomic factors, and power grid operation factors.

[0067] The environmental factors, socio-economic factors, and power grid operation factors include temperature, humidity, economic growth, power grid compliance, and power supply status.

[0068] When the conditional independence obtained from the independence test is close to 0, X and Y are considered independent under the given conditions, which helps to remove indirect influencing variables. Next, a linear non-Gaussian model is used to further infer the direction of causal relationships, and optimization is performed in conjunction with causal intervention strategies. The specific formula for the independence test is as follows:

[0069] CI(X,Y|Z)=P(X|Y,Z)-P(X|Z)

[0070] Where X and Z represent influencing variables, and Y represents the target variable. When the value of CI(X,Y|Z) is approximately 0, it is assumed that variables X and Y are independent given Z.

[0071] The obtained causal relationship is further inferred using a linear non-Gaussian model, and the specific regression model formula is expressed as follows:

[0072] Y=βX+∈

[0073] Where ∈ represents the error term, and β represents the causal effect coefficient. If the error term ∈ is not independent of the influencing variable X, then X→Y is the causal direction.

[0074] By searching for the optimal structure in all possible causal graphs using a greedy equivalence search, a causal graph that maximizes the causal structure is found. The formula for optimizing the causal structure is expressed as:

[0075] Among them, P(D) i |G) represents data D i The probability D under the causal graph G i Let represent the i-th data point, G represent the causal graph, and argmax represent the argmax value. G The expression represents selecting the G among all causal graphs G that maximizes the subsequent expression.

[0076] In practical data analysis, correlation between variables is not equivalent to causation. Spurious correlations are usually caused by underlying common causes or other external factors, rather than direct causal effects, and need to be eliminated from the obtained causal relationships. First, define the intervention operation, expressed by the following formula:

[0077] P(Y|do(X))

[0078] Where do(X) represents the human control of X, and the expected value of Y after a given intervention is obtained through the intervention operation formula.

[0079] To calculate causal influence, we obtain the causal influence with low correlation. The specific formula for calculating causal influence is as follows:

[0080] ΔY = E[Y|do(X1)] - E[Y|do(X2)]

[0081] Where ΔY represents the causal change effect of variable Y, do(X) represents the intervention operation in causal inference, and E represents the expectation symbol E[Y|do(X)] represents the conditional expectation.

[0082] The Pearson correlation coefficient was calculated, and variables with low correlation but no causal relationship were removed based on the obtained Pearson correlation coefficient. Researchers established a correlation threshold based on data analysis and experience, and then removed data with low correlation.

[0083] The variables with causal relationships, after eliminating spurious correlations, are scored. The importance scoring formula is expressed as follows:

[0084] S(X)=w1·|ρ(X,Y)|+w2·I(Y|do(X))

[0085] Where S(X) represents the importance score of variable X, w1 and w2 represent weight parameters, and I(Y|do(X)) represents the amount of information about how variable X affects Y through causal intervention. The top k variables with the highest S(X) are selected to form the variable dataset X. * X * ={X1,X2,...,X k}

[0086] A new coefficient is introduced to adjust the tolerance for scheduling errors. A deep learning model is constructed to enhance the model's flexibility. The deep learning model optimizes power scheduling based on power demand values, minimizing power supply costs while reducing power demand prediction errors. By combining power supply costs with scheduling error penalties and scheduling flexibility control, power demand scheduling optimization is achieved. The specific formulas for power supply costs and scheduling error penalties are as follows:

[0087] The main part that deals with the cost of power supply and the penalty for dispatching errors ensures that the objective function is minimized by taking into account both the actual cost of power supply and a reasonable penalty for forecasting errors.

[0088] The specific formula for scheduling flexibility control is expressed as follows:

[0089] F=θ·(P a (t)-P p (t)))

[0090] The dispatch flexibility control section introduces a dispatch flexibility control coefficient. By adjusting this coefficient, the impact of dispatch errors on the objective function is controlled, thereby further optimizing the power dispatch process.

[0091] The specific formula for calculating the scheduling flexibility control coefficient is as follows:

[0092]

[0093] By weighting historical dispatch data to calculate the dispatch flexibility control coefficient θ, the relationship between forecast error and power supply cost can be reflected. Introducing the flexibility control coefficient can make the dispatch optimization process more flexible in scenarios where power demand fluctuates greatly.

[0094] The specific formula for optimizing power demand dispatch is expressed as follows:

[0095] L = E + F

[0096] Where L represents the total loss function of power dispatch optimization, P a (t) represents the actual power supply at time t, P p (t) represents the predicted electricity demand at time t, c t d represents the cost of electricity supply at time t. t θ represents the error penalty coefficient at time t, and θ represents the scheduling flexibility control coefficient.

[0097] A deep learning model is constructed using a Long Short-Term Memory (LSTM) network. The causal inference results are then combined with the deep learning model to capture time-series patterns in electricity demand, enhancing the model's predictive ability by leveraging causal relationships. Causal inference identifies which causal variables have a significant impact on electricity demand, and causal intervention quantifies this impact. These causal variables are then used as inputs to perform time-series predictions in conjunction with the deep learning model. The specific formula for calculating causal variables is as follows:

[0098]

[0099] Here, A represents the weighted component of causal variables and historical data. It captures the impact of causal relationships on electricity demand by weighting the causal variables. The output of the causal inference model is combined with the time-series learning results of the LSTM network to further optimize the electricity forecasting process. The specific forecasting optimization formula is expressed as:

[0100] B = γ·f LSTM X * δ·P c X *

[0101] Where B represents the output of the LSTM and causal inference model. This part mainly combines the output of the causal inference model with the time-series learning results of the LSTM network to further optimize the prediction of electricity demand. The specific formula of the combined model is expressed as follows:

[0102] Y p,t =A+B

[0103] Among them, Y p,t X represents the predicted electricity demand at time t. c,i,t-kLet α represent the data of the i-th causal variable at time tk. i β represents the coefficient of influence of the i-th causal variable on electricity demand. k f represents the weighting coefficient for time tk in the historical data. LSTM P represents the output of the LSTM model. c This represents the output of the causal inference model. This combined model extracts time-series features from historical data and adjusts the input variables based on causal relationships. This makes the predictions more focused on actual causal relationships, enhancing the model's interpretability and accuracy.

[0104] To ensure the accuracy and feasibility of power dispatch optimization and to guarantee its applicability to actual power system dispatch problems, a nonlinear programming method combined with gradient descent is used to solve the power dispatch optimization problem. Nonlinear programming is suitable for optimization problems involving quadratic terms and nonlinear constraints. Since the total loss function includes the squared term of the dispatch error, nonlinear programming can effectively handle these nonlinear terms and find the optimal power dispatch scheme. The constraints are defined as capacity limitations of power supply and system load balance constraints, specifically expressed as follows:

[0105]

[0106] Among them, P min P represents the minimum power supply at time t. max P min This represents the maximum power supply at time t.

[0107]

[0108] This constraint ensures that actual electricity supply matches forecasted electricity demand.

[0109] Gradient descent is a commonly used optimization algorithm, especially when the objective function is continuously differentiable. Gradient descent can converge quickly and find the optimal solution. For nonlinear optimization in power dispatching problems, gradient descent can effectively adjust decision variables and reduce the objective function value. First, the gradient of the loss function with respect to the decision variables is calculated. The specific formula for calculating the gradient is as follows:

[0110]

[0111] Partial derivative of the force supply.

[0112] The power supply value is adjusted by updating the rules. The specific formula for the rule update is as follows:

[0113]

[0114] Where η represents the learning rate, Pa (t) old P represents the power supply value in the previous iteration. a (t) new This indicates the updated power supply value.

[0115] Power dispatch is implemented based on the optimized power supply plan to ensure that the difference between actual power supply and predicted power demand is minimized. After dispatch is implemented, the actual operation of the power system is continuously monitored. If significant errors occur, the dispatch strategy is adjusted based on real-time feedback.

[0116] Further explanation is needed regarding the real-time feedback information adjustment scheduling strategy, which includes real-time monitoring of the deviation between the actual power supply and the predicted demand value of the power system. When the percentage deviation for a consecutive preset number of time periods exceeds the threshold of the corresponding area type, a dynamic adjustment mechanism is triggered. The current real-time data is re-verified, and linear interpolation is used to perform time synchronization calibration on multi-source data. Long short-term memory networks are used to reconstruct and verify abnormal power supply data. Based on the re-verified data, the causal relationship graph is updated, and new causal variables are selected through causal discovery algorithms. The updated causal variables are input into the demand forecasting model to generate a corrected power demand forecast value. The gradient descent method is used to optimize the scheduling of the corrected forecast value, generating a new power supply plan. The optimized supply plan is then distributed to the distribution automation equipment for execution through the power system communication protocol.

[0117] Example 2 is the second embodiment of the present invention:

[0118] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0119] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0120] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0121] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0122] Example 3 is the third embodiment of the present invention. This embodiment provides an artificial intelligence-based power demand data analysis system, including a multi-source data governance module, a causal inference engine module, a time-series-causal fusion prediction module, and an adaptive scheduling optimization module.

[0123] The multi-source data governance module uses a time alignment engine and an anomaly detection unit to achieve cross-source data synchronization and quality improvement.

[0124] The causal inference engine module is based on a hybrid architecture of PC algorithm and linear non-Gaussian model. It performs variable independence test and causal direction inference, and combines greedy search to generate the optimal causal graph to eliminate spurious correlation interference.

[0125] The time-series-causal fusion prediction module extracts historical load time-series features through an LSTM network and dynamically weights and fuses them with the key variable weights output by causal inference to form an interpretable composite prediction model.

[0126] The adaptive scheduling optimization module integrates a nonlinear programming solver and a real-time feedback mechanism. It dynamically adjusts the power supply strategy based on the prediction results, it iteratively optimizes the supply and demand matching through the gradient descent algorithm, and triggers the system's self-repair process based on deviation monitoring.

[0127] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based method for analyzing electricity demand data, characterized by: include, Collect heterogeneous data from multiple sources and preprocess the data; Based on the preprocessed data, a causal discovery algorithm is used to analyze the causal relationships between multi-source data, remove irrelevant factors, select the final variables, and output the causal inference results. Build a deep learning model, optimize it based on causal reasoning results, and output prediction results; Optimize power dispatching strategies based on forecast results; The analysis of causal relationships among multi-source data using causal discovery algorithms includes performing variable independence tests using PC algorithms and removing redundant variables when the conditional independence value is below a threshold. A linear non-Gaussian model is used to infer causal directions, and a greedy search is combined to optimize the causal structure. The optimized power dispatch strategy includes monitoring the continuous deviation between the actual power supply and the predicted value, triggering multi-source data recalibration and cause-effect graph update, using LSTM to verify abnormal data and then generating corrected predictions, optimizing through gradient descent and issuing new dispatch instructions. The analysis of causal relationships among multi-source data using causal discovery algorithms based on preprocessed data includes, Based on the preprocessed multi-source heterogeneous data, candidate variables related to electricity demand are initially selected through physical mechanisms. The independence of causal relationships is then tested using the PC algorithm to eliminate redundant variables. The candidate variables include environmental variables and power grid operation variables; Environmental variables, including power grid operation variables, include temperature, humidity, power grid load, and power supply status; When the value of the conditional independence test is less than the set threshold, the current events are independent under the given conditions. The direction of causal relationship is inferred using a linear non-Gaussian model, and optimization is carried out in combination with causal intervention strategies. The optimal causal structure is searched among all causal relationships using a greedy equivalence search, outputting a maximally optimized causal graph. The formula for optimizing the causal structure is expressed as: in, Representing data The probability under the causal graph G, Indicates the first One data point, Represents a cause-and-effect graph. Indicates all causal graphs In the middle, choose the one that maximizes the subsequent expression. ; The process of eliminating irrelevant factors to select the final variables includes, Use defined intervention operations to remove spurious correlation factors, determine the expected value after intervention through intervention operations, calculate the influence between causality, output the causal influence with low correlation, remove variables with low correlation but no causal relationship through Pearson correlation coefficient, and set a correlation threshold to remove data with correlation below the threshold. The variables with causal relationships, after eliminating spurious correlations, are scored. The importance scoring formula is expressed as follows: in, Representing variables Importance rating , Represents the weight parameters. Representing variables Influence through causal intervention Information volume, selection The top k variables constitute the variable dataset. , .

2. The method for analyzing electricity demand data based on artificial intelligence as described in claim 1, characterized in that: The process of collecting multi-source heterogeneous data and preprocessing the data includes, Load data is collected through industrial communication protocols, received in real time using a distributed stream processing platform, stored in a time-series database, and the electricity consumption growth rate of industrial and commercial users is obtained through power company reports. Power grid status data is obtained from the SCADA system via the IEC61850 communication protocol, and power monitoring is performed using a synchronous phasor measurement device. The preprocessing includes synchronizing the time of multi-source data to a unified time step using time alignment technology, synchronizing the time of low-frequency data using linear interpolation to ensure consistent time steps across data sources, processing outliers based on the statistical 3σ criterion, and normalizing each data point using Min-Max.

3. The method for analyzing electricity demand data based on artificial intelligence as described in claim 2, characterized in that: The deep learning model is constructed and optimized based on causal inference results, and the output prediction results include... A new coefficient is introduced to adjust the tolerance for scheduling errors. A deep learning model is constructed, which optimizes power scheduling based on power demand values, minimizes power supply costs, and reduces power demand prediction errors. By combining power supply costs with scheduling error penalty and scheduling flexibility control, power demand scheduling is optimized. A deep learning model is constructed using a temporal neural network model. The causal inference results obtained are combined with the deep learning model to capture the time series patterns of electricity demand and enhance the predictive ability of the deep learning model by utilizing causal relationships.

4. The method for analyzing electricity demand data based on artificial intelligence as described in claim 3, characterized in that: The optimization of power dispatch strategy based on prediction results includes: The power dispatch optimization problem is solved by combining nonlinear programming with gradient descent. The constraints are set as capacity limits of power supply and balance constraints of system load. Power dispatch is implemented according to the optimized power supply plan so that the difference between actual power supply and predicted power demand approaches 0. After the dispatch is implemented, the actual operation of the power system is continuously monitored. If the error exceeds the set threshold, the dispatch strategy is adjusted based on real-time feedback information.

5. The method for analyzing electricity demand data based on artificial intelligence as described in claim 4, characterized in that: The adjustment of the scheduling strategy based on real-time feedback information includes... The system monitors the deviation between the actual power supply and the predicted demand in real time. When the percentage deviation exceeds the threshold for the corresponding area type for a number of consecutive preset time periods, a dynamic adjustment mechanism is triggered. The current real-time data is re-verified, and the time synchronization calibration of multi-source data is performed using linear interpolation. The abnormal power supply data is reconstructed and verified using a long short-term memory network. The causal relationship graph is updated based on the re-verified data. New causal variables are selected through causal discovery algorithms. The updated causal variables are input into the demand forecasting model to generate revised electricity demand forecasts. The gradient descent method is used to optimize the scheduling of the revised forecasts, generating new power supply schemes. The optimized supply schemes are then distributed to distribution automation equipment for execution through the power system communication protocol.

6. A system employing the artificial intelligence-based power demand data analysis method as described in any one of claims 1 to 5, characterized in that: It includes a multi-source data governance module, a causal inference engine module, a time-series-causal fusion prediction module, and an adaptive scheduling optimization module; The multi-source data governance module uses a time alignment engine and an anomaly detection unit to achieve cross-source data synchronization and quality improvement. The causal inference engine module is based on a hybrid architecture of PC algorithm and linear non-Gaussian model. It performs variable independence test and causal direction inference, and combines greedy search to generate the optimal causal graph to eliminate spurious correlation interference. The time-series-causal fusion prediction module extracts historical load time-series features through an LSTM network and dynamically weights and fuses them with the key variable weights output by causal inference to form an interpretable composite prediction model. The adaptive scheduling optimization module integrates a nonlinear programming solver and a real-time feedback mechanism. It dynamically adjusts the power supply strategy based on the prediction results, iteratively optimizes the supply and demand matching through the gradient descent algorithm, and triggers the system self-repair process based on deviation monitoring.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the artificial intelligence-based power demand data analysis method as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based power demand data analysis method as described in any one of claims 1 to 5.

Images