A method and system for predicting power consumption in a smart park based on artificial intelligence

Abstract

The present invention discloses a method and system for predicting power consumption of a smart park based on artificial intelligence, which relates to the field of power prediction. The method includes: collecting multi-source heterogeneous energy consumption data of the smart park, pre-processing the multi-source heterogeneous energy consumption data, and building a migration sample library based on the pre-processed multi-source heterogeneous energy consumption data; building a spatiotemporal energy consumption prediction model that integrates time and space dimensions, and using sample data in the migration sample library to train the spatiotemporal energy consumption prediction model, predicting the power consumption of the smart park in the future time period through the trained spatiotemporal energy consumption prediction model to obtain a power consumption prediction result; building an early warning and control mechanism based on the power consumption prediction result of the smart park, and realizing power resource management of the smart park through the early warning and control mechanism. The present invention can accurately predict the energy consumption change trend of the park's electricity in different time and space dimensions, and discover potential energy consumption anomalies in advance.

Description

Technical Field

[0001] The present invention relates to the field of power forecasting technology, and in particular to an artificial intelligence-based power consumption forecasting method and system for a smart park. Background Art

[0002] With the rapid advancement of information technology, the widespread application of technologies such as the Internet of Things (IoT), big data, artificial intelligence (AI), and cloud computing is driving the transformation of traditional industrial parks into smart parks. By integrating advanced communications and information technology, smart parks optimize resource allocation, improve operational efficiency, and contribute positively to the sustainable development of cities. However, with the continuous expansion of industrial parks and the increasing diversification of power demands, the management and forecasting of power consumption has become a major challenge that needs to be addressed in the construction of smart parks. In particular, the temporal and spatial variations in power consumption make traditional power management methods difficult to effectively address the increasingly complex energy conservation needs of industrial parks.

[0003] Traditional power consumption management methods are mostly based on static statistical models or time series analysis. These methods focus on processing historical power data but ignore the spatial variations and temporal fluctuations in power consumption within a park. In particular, with the increasing diversity of buildings, facilities, and equipment within a park, the spatial distribution and temporal variations of power demand exhibit more complex spatiotemporal dynamics. Traditional methods often struggle to achieve accurate forecasts when addressing these spatiotemporal variations.

[0004] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides an artificial intelligence-based smart park power consumption prediction method and system, which has the advantage of combining the spatial differences and temporal fluctuation characteristics of power consumption to predict the park's power consumption, thereby solving the problem of spatial differences and temporal fluctuation characteristics of power consumption within the park.

[0006] In order to achieve the advantages of combining the spatial differences and temporal fluctuation characteristics of power consumption to predict the power consumption of the park, the specific technical solutions adopted by the present invention are as follows:

[0007] According to one aspect of the present invention, a method for predicting power consumption of a smart park based on artificial intelligence is provided, the method comprising:

[0008] Collect and pre-process the multi-source heterogeneous energy consumption data of the smart park, and build a migration sample library based on the pre-processed multi-source heterogeneous energy consumption data;

[0009] Construct a spatiotemporal energy consumption prediction model that integrates time and space dimensions, and use the sample data in the migration sample library to train the spatiotemporal energy consumption prediction model. After training, the spatiotemporal energy consumption prediction model is used to predict the power consumption of the smart park in the future time period to obtain the power consumption prediction results.

[0010] Based on the power consumption prediction results of the smart park, an early warning and control mechanism is established to achieve power resource management of the smart park through the early warning and control mechanism.

[0011] According to another aspect of the present invention, there is also provided a smart park power consumption prediction system based on artificial intelligence, the system comprising:

[0012] The sample library construction module is used to collect and preprocess the multi-source heterogeneous energy consumption data of the smart park, and build a migration sample library based on the preprocessed multi-source heterogeneous energy consumption data;

[0013] The power consumption prediction module is used to build a spatiotemporal energy consumption prediction model that integrates the time and space dimensions. It uses the sample data in the migration sample library to train the spatiotemporal energy consumption prediction model. The trained spatiotemporal energy consumption prediction model is used to predict the power consumption of the smart park in the future time period to obtain the power consumption prediction results.

[0014] The power resource management module is used to build an early warning and control mechanism based on the power consumption prediction results of the smart park, and realize the power resource management of the smart park through the early warning and control mechanism.

[0015] Compared with the existing technology, the present invention provides an artificial intelligence-based smart park power consumption prediction method and system, which has the following beneficial effects: the present invention can accurately predict the energy consumption change trend of the park's electricity in different time and space dimensions, discover potential energy consumption anomalies in advance and make effective adjustments, thereby improving the accuracy of energy consumption management, and enhancing the stability and reliability of the system. At the same time, it dynamically optimizes equipment operation based on real-time data and historical records to ensure that energy consumption is always optimized in a changing environment and demand, thereby effectively improving energy utilization efficiency, reducing carbon emissions, and providing a solid technical guarantee for achieving green energy development goals. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1is a flow chart of a method for predicting power consumption of a smart park based on artificial intelligence according to an embodiment of the present invention;

[0018] Figure 2 This is a principle block diagram of an artificial intelligence-based smart park power consumption prediction system according to an embodiment of the present invention.

[0019] In the picture:

[0020] 1. Sample library construction module; 2. Power energy consumption prediction module; 3. Power resource management module. DETAILED DESCRIPTION

[0021] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and the advantages of the present invention.

[0022] According to an embodiment of the present invention, a method and system for predicting power consumption of a smart park based on artificial intelligence are provided.

[0023] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to an embodiment of the present invention, a method for predicting power consumption of a smart park based on artificial intelligence includes:

[0024] S1. Collect multi-source heterogeneous energy consumption data of the smart park, pre-process the multi-source heterogeneous energy consumption data, and build a migration sample library based on the pre-processed multi-source heterogeneous energy consumption data.

[0025] The multi-source heterogeneous energy consumption data of the smart park is collected and pre-processed. The migration sample library is constructed based on the pre-processed multi-source heterogeneous energy consumption data, including:

[0026] Use IoT sensor devices to collect multi-source heterogeneous energy consumption data of smart parks, and multi-source heterogeneous energy consumption data includes power equipment operation data, environmental variable data, and personnel activity data;

[0027] It should be noted that the present invention fully considers the multi-source data characteristics of the smart park, and comprehensively collects data in multiple dimensions such as IoT device operation data, environmental information, and historical load records. Among them, the equipment operation data covers basic parameters such as power consumption, voltage, and current, which truly reflects the power consumption patterns of various types of equipment within the park; environmental information covers external environmental factors such as temperature and humidity, which helps to deeply analyze the potential impact of environmental changes on electricity demand; historical load records are used to reveal the periodicity and trend characteristics of energy consumption. In addition, the present invention also incorporates personnel activity data to reflect the spatiotemporal correlation between the density of human traffic and electricity consumption in the park. By integrating and uniformly modeling these multi-dimensional data, a high-quality feature matrix is ​​provided for the prediction model, which significantly enhances the adaptability and prediction accuracy of the model.

[0028] An outlier detection algorithm is used to remove abnormal data caused by environmental interference factors in multi-source heterogeneous energy consumption data, and a linear interpolation algorithm is used to fill in missing data in multi-source heterogeneous energy consumption data;

[0029] Perform normalization and noise reduction on the multi-source heterogeneous energy consumption data in sequence to eliminate data dimension differences and high-frequency noise interference, and obtain pre-processed multi-source heterogeneous energy consumption data;

[0030] Reconstruct the pre-processed multi-source heterogeneous energy consumption data to form a spatiotemporal data matrix that reflects the power consumption patterns of the smart park. Calculate the power consumption of the smart park samples based on the spatiotemporal data matrix, and integrate the calculation results to generate core feature sample data.

[0031] It should be noted that the present invention implements a series of cleaning, conversion, and reconstruction processes for the collected multi-source heterogeneous IoT data. Specifically, outlier removal and missing value interpolation methods are used to address data quality issues, while normalization and noise reduction techniques are employed to eliminate data dimensionality differences and high-frequency noise interference. Ultimately, through reconstruction of the temporal and spatial dimensions, a high-dimensional spatiotemporal data matrix is ​​successfully constructed, ensuring that the model accurately captures the dynamic changes in electricity consumption.

[0032] A migration sample library is constructed based on core feature sample data, and cross-level sample migration is triggered in the migration sample library according to predefined power consumption forecast requirements.

[0033] It should be noted that the spatiotemporal energy consumption prediction model of the present invention is designed to predict the power consumption information of each area (spatial dimension) and different time periods (temporal dimension) within the smart park. The specific steps for constructing the sample library required for model training are as follows:

[0034] 1. The primary task in building the sample library is to collect comprehensive and multi-dimensional power consumption data. By deploying smart meters, environmental sensors, and crowd monitoring devices throughout the park, we capture operational parameters such as power consumption, voltage, and current. We also collect data on environmental variables such as temperature, humidity, and light intensity. Furthermore, by connecting to the weather forecast system and the park's historical data system, we obtain information on weather changes and historical load trends. All collected data is uploaded and integrated in real time via the IoT platform, providing multi-source, heterogeneous data support for subsequent analysis.

[0035] 2. To ensure data quality, comprehensive preprocessing of the raw data is required before building the sample library. First, outlier detection techniques are used to remove abnormal data caused by equipment failure or environmental interference, and linear interpolation is used to fill in missing data. Second, the data is normalized to ensure uniformity across different dimensions. Noisy data is cleaned using methods such as low-pass filtering or wavelet noise reduction. Furthermore, the data is segmented and labeled according to time and space dimensions (e.g., time periods of 15 minutes, 1 hour, 1 day, and 1 week, and spatial divisions into equipment levels and functional zoning) to form a preprocessed dataset that reflects spatiotemporal characteristics.

[0036] 3. After data preprocessing, sample energy consumption for various devices and areas must be calculated to generate core feature sample data for predictive model training. Real-time power and cumulative energy consumption are calculated based on collected data such as device power consumption, voltage, and current. Furthermore, energy consumption data is corrected using a multivariate weighted approach, taking into account environmental conditions (such as the impact of temperature on air conditioning energy consumption) and human activity intensity (such as increased equipment load during periods of high traffic) to produce energy consumption indicators that more closely reflect actual electricity usage scenarios. The results of these energy consumption calculations will serve as a key component of the subsequent sample library, providing accurate numerical features for model input.

[0037] 4. Based on the needs of smart park power consumption forecasting, a multi-layered and multifunctional sample library was constructed. In terms of time, the sample library has sub-libraries based on short-term, medium-term, and long-term needs to meet the requirements of different forecasting tasks. In terms of space, it covers power consumption data for the entire park, building areas, and individual devices (equipment categories). Furthermore, the sample library utilizes a structured storage method, storing energy consumption data, environmental data, and personnel activity data in layers. This data is then efficiently retrieved and managed through a time-series database. The sample library also incorporates a dynamic update mechanism, regularly updating and supplementing data based on real-time data. Sample annotations are optimized based on user feedback to improve data timeliness and accuracy.

[0038] 5. After the sample library is constructed, the labeled data is used to train and validate the AI ​​spatiotemporal prediction model. Stratified sampling is used to divide the data into training, validation, and test sets to ensure representative data distribution. During model training, time series features are extracted using a modified long short-term memory (LSTM) network to capture long-term and short-term dependencies in time series data. During the validation phase, cross-validation is used to evaluate the model's generalization capabilities, and error analysis metrics (such as mean squared error and mean absolute error) are used to measure the model's prediction accuracy. The validation results will further guide the optimization of the sample library and the adjustment of model parameters to ensure the model's ability to effectively adapt to the complex scenarios of smart park power consumption.

[0039] Among them, a migration sample library is constructed based on core feature sample data, and cross-level sample migration is triggered in the migration sample library according to predefined power consumption forecast requirements, including:

[0040] Based on the core feature sample data, an initial sample library is constructed and the core feature sample data is classified and stored. According to the classified sample categories, the initial sample library is divided into several sub-libraries according to the time dimension and the space dimension;

[0041] Build a hierarchical migration tree in each sub-database and calculate the migration offset between the current hierarchical migration tree and each of the remaining hierarchical migration trees.

[0042] It's important to note that a hierarchical migration tree is a tree structure used to represent the migration of data or tasks from one level to another in a multi-level system. In the hierarchical migration tree, each node represents a level in the multi-level system, and each edge represents the data or task transfer relationship between different levels.

[0043] Calculating the migration offset between the current level migration tree and each of the remaining level migration trees includes:

[0044] Extract the target sample set from the core feature sample data based on the spatiotemporal data matrix, calculate the total score value of each gene sample in the target sample set, and sort the total score value of each gene sample to obtain the gene sequence;

[0045] Extract a non-target sample set from the core feature sample data, calculate the total score value of each non-gene sample in the non-target sample set, and sort the total score value of each non-gene sample to obtain a non-gene sequence;

[0046] It should be noted that extracting the target sample set from the core feature sample data and calculating the total score of the target sample includes: constructing a spatiotemporal data matrix, where each row of the spatiotemporal data matrix represents a sample and each column represents a feature (such as an environmental factor or biomarker); selecting core features related to the target class samples in the spatiotemporal data matrix; calculating the total score of the target sample based on the core features; standardizing the score of each target sample to make the score comparable; and sorting the scores of all target samples to obtain the ranking of the gene samples according to the scores.

[0047] The association analysis algorithm is used to calculate the sorting difference between gene sequences and non-gene sequences in the gene dimension, and the sorting difference is used as the migration offset between the current level migration tree and each of the remaining level migration trees.

[0048] It should be noted that the association analysis algorithm is a gene ranking difference analysis algorithm, and the calculation formula for the ranking difference between gene sequences and non-gene sequences in the gene dimension is:

[0049] ;

[0050] Where, WGS Indicates the difference in the order of gene sequences and non-gene sequences in the gene dimension; Indicates the first i The total score of the gene samples; Indicates the first i The total score of non-genetic samples; α represents the attenuation coefficient, which is used to control the weight of high-ranking genes; n represents the total number of gene samples; Indicates the first i The sequencing position of each gene sample; Indicates the first j The sequencing position of each gene sample; e Represents a constant.

[0051] Based on the sample migration offset, the samples covered by the rule features with the highest confidence are migrated from the current sub-library to the adjacent sub-library. After the migration is completed, the new rule features of the remaining samples are calculated and used as the basis for a new round of iterative migration until the number of remaining samples is empty or less than the set threshold. Before the iterative migration of samples, a hierarchical migration tree classifier is constructed, and the priority order of sample migration is formulated according to the predefined power consumption prediction requirements. The hierarchical migration tree classifier triggers the migration of samples in the sub-library according to the priority order.

[0052] S2. Construct a spatiotemporal energy consumption prediction model that integrates the time dimension and the space dimension, and use the sample data in the migration sample library to train the spatiotemporal energy consumption prediction model. After training, the spatiotemporal energy consumption prediction model is used to predict the power consumption of the smart park in the future time period to obtain the power consumption prediction result.

[0053] It should be noted that to achieve high-precision forecasting, this paper constructs a spatiotemporal prediction model for campus energy consumption that integrates time series analysis and spatial modeling techniques. This model uses a modified long short-term memory (LSTM) network to extract the temporal characteristics of device power consumption, accurately capturing its cyclical fluctuations and long-term trends; thereby achieving a combined spatiotemporal forecast of campus power consumption. The model incorporates a multi-factor weight fusion mechanism, comprehensively considering factors such as environmental variables, device parameters, and historical load data, ensuring its flexibility in adapting to various complex scenarios.

[0054] Among them, a spatiotemporal energy consumption prediction model that integrates the time dimension and the space dimension is constructed, and the spatiotemporal energy consumption prediction model is trained using the sample data in the migration sample library. The trained spatiotemporal energy consumption prediction model is used to predict the power consumption of the smart park in the future time period. The power consumption prediction results include:

[0055] Based on the improved long short-term memory network and superposition theorem layer network, a spatiotemporal prediction model for park power consumption is constructed;

[0056] Among them, the time series characteristics of power consumption are extracted through the improved long short-term memory network to capture the periodic change trend of power consumption, and the interpretable energy consumption activation function of the theorem layer network is superimposed to reduce the number of parameters in the spatiotemporal prediction model of energy consumption.

[0057] It should be noted that the present invention applies a customized intelligent AI model to energy consumption prediction for smart parks. During the iterative optimization process of the model, an online learning mechanism is introduced into the iterative training of the model, enabling the model to receive real-time data from Internet of Things (IoT) sensors in real time and automatically adjust weight parameters through training. This self-optimization process ensures the real-time and adaptability of the model, enabling it to dynamically adjust according to changes in the park environment, thereby continuously maintaining efficient energy consumption prediction capabilities. This dynamic learning model further improves the model's sensitivity to changes in park operating scenarios and promotes the continuous optimization and development of energy consumption prediction technology, which specifically includes:

[0058] Model Structure: This intelligent AI model utilizes an improved LSTM architecture to enhance the spatiotemporal modeling capabilities for smart park energy consumption forecasting. By combining the temporal characteristics of time series data with the spatial characteristics of application scenarios, the model accurately predicts energy consumption levels. The model includes the following innovative structures.

[0059] Improved Scalar Long Short-Term Memory (S-LSTM): This model innovatively introduces exponential gating and memory mixing mechanisms, enhancing the performance of traditional LSTM networks when processing time series data. S-LSTM not only supports multiple memory cells but also implements memory cell mixing through recursive connections. This enables the model to effectively extract complex patterns and make accurate predictions for tasks that require long-term state tracking.

[0060] In smart campus energy consumption forecasting, electricity usage data is a typical time series scalar data. The S-LSTM network accurately captures the temporal dependencies of this data, thereby improving the accuracy of energy consumption forecasts. By introducing exponential gating, the model can flexibly adjust its memory and forgetting decisions, ensuring adaptability to dynamic temporal changes. The introduction of a memory hybrid mechanism enables multiple memory units to work together, further enhancing the ability to extract complex spatiotemporal patterns.

[0061] Superposition Theorem Layer Network (Kolmogorov-Arnold Network): Unlike the fixed activation functions used in traditional neural networks, the Superposition Theorem model introduces learnable activation functions at the edge of the network. This invention designs specialized energy consumption activation functions for application scenarios involving multi-source heterogeneous energy consumption data. These activation functions are typically parameterized using nonlinear energy consumption piecewise functions, providing a high degree of flexibility for the model. Each energy consumption weight parameter is replaced with a single-variable function, allowing the model to express complex functions with fewer parameters. This not only improves the model's flexibility but also enhances its interpretability. This approach significantly improves the model's computational efficiency when faced with complex tasks.

[0062] In this model, the superposition theorem layer significantly improves the performance of smart park energy consumption forecasting by accurately modeling spatial dependencies. This layer deeply processes input data through recursive connections and nonlinear transformations, effectively extracting high-dimensional spatial features. Compared to traditional multilayer perceptrons (MLPs), the superposition theorem layer not only significantly reduces the number of model parameters but also effectively reduces system storage requirements, thereby optimizing computational efficiency.

[0063] The AI ​​model of the present invention fully leverages the advantages of the scalar long short-term memory (S-LSTM) module and the Kolmogorov–Arnold Network (KAN) layer by combining them to form a powerful spatiotemporal prediction system. In the S-LSTM module, the exponential gating and memory mixing mechanisms enhance the modeling capabilities of long-term and short-term dependencies in multi-source energy consumption scalar time series data, enabling the model to accurately predict the dynamic changes in campus energy consumption. The KAN layer, through its new network perspective and spatial modeling capabilities, effectively captures the complex spatial dependencies between campus equipment and regions, further improving the model's performance in spatial feature learning. The KAN layer also further optimizes system storage and computing efficiency by reducing the number of parameters. The combination of the three enables this model to not only have powerful spatiotemporal modeling capabilities, but also provide accurate energy consumption forecasts in real time in complex and dynamic campus environments.

[0064] Determine whether there is a prediction bias in the power energy consumption forecast result. If there is no prediction bias, the power energy consumption forecast result output by the energy consumption spatiotemporal forecast model is used as the final result. Otherwise, the deviation in the power energy consumption forecast result is corrected based on the potential superposition effect between sample features to obtain a corrected power energy consumption forecast result.

[0065] Among them, based on the potential superposition effect between sample characteristics, the deviation in the power consumption forecast result is corrected, and the corrected power consumption forecast result includes:

[0066] Extracting sample features after classification of core feature sample data, and the sample features include power equipment operation features, environmental variable features, and personnel activity features;

[0067] The operating status of adjacent power equipment is aggregated based on the operating characteristics of the power equipment, and the power equipment jump deviation existing in the start and stop of the power equipment in the aggregation of the operating status of adjacent power equipment is analyzed.

[0068] It should be noted that for the quantification of jump deviation, the improved Hampel filter algorithm is used to identify abnormal residuals in the operating state residuals, and the jump deviation index is defined as:

[0069] ;

[0070] Where, D k Indicates power equipment k The jump deviation index, Indicates power equipment k When a start-stop event of a power device is detected (determined by the power change rate threshold), the operating state residuals of all adjacent devices within 5 seconds before and after the event are extracted. , Indicates power equipmentk The measured operating status within the start-stop time window (such as the instantaneous current value), Represents the power equipment predicted by the LSTM model k In the steady-state operation state, Δ represents the time window (when the start-stop event of the power equipment is detected, 5 seconds are taken before and after the event, forming a time window with a total length of 10 seconds). The model input is the current and voltage series of the equipment during the historical steady-state period without mutations. The model output is the predicted steady-state operating value at the current moment. ΔS k When ΔS is greater than 0, the actual operating status is higher than expected, which may be due to a sudden increase in load or equipment abnormality. k When <0, the actual operating status is lower than expected, which may be due to equipment shutdown or power outage; μ and σ Indicates the mean and standard deviation of the residuals of adjacent devices. When it exceeds 3 σ Finally, through the back-propagation correction mechanism, the jump deviation is fed back to the device control strategy to dynamically adjust the protection setting or power allocation weight of adjacent devices.

[0071] The improved particle filter algorithm is used to simulate the superposition effect between the operating characteristics of power equipment and the characteristics of environmental variables, and to capture the mutual adaptation deviation between the operation of power equipment and environmental factors.

[0072] Among them, the improved particle filter algorithm is used to simulate the superposition effect between the operating characteristics of power equipment and the characteristics of environmental variables, and to capture the mutual adaptation deviation between the operation of power equipment and environmental factors, including:

[0073] Randomly generate several particles based on the operating characteristics of power equipment and environmental variables and assign initial weights. Construct a state transition equation to analyze the current particle state quantity, and predict the particle state quantity at the next time point based on a first-order Markov process.

[0074] Compare the difference between the particle state at the next time point and the current particle state with a preset threshold, update the particle weight based on the comparison result, and calculate the adaptation relationship between the power equipment operation characteristics and the environmental variable characteristics based on the updated particle weight;

[0075] Particle boosting is reconstructed based on the adaptation relationship, and a particle state estimator is constructed using an improved particle filter algorithm. The source intensity distribution during the particle reconstruction boosting process is identified through the particle state estimator, and the mutual adaptation deviation between the operation of power equipment and environmental factors is identified based on the particle source intensity distribution.

[0076] It should be noted that the improved particle filter algorithm introduces a comparison between the difference between the particle state and the current state and a preset threshold when updating the particle weight. This mechanism can adjust the particle weight more sensitively, especially in a rapidly changing environment, and can more accurately reflect changes in the device and environmental status.

[0077] Among them, particle boosting is reconstructed based on the adaptation relationship, and a particle state estimator is constructed using an improved particle filter algorithm. The particle state estimator is used to identify the source intensity distribution during the particle reconstruction boosting process. Based on the particle source intensity distribution, the mutual adaptation deviation between the operation of power equipment and environmental factors is identified. The following are included:

[0078] The fusion particles reconstruct the particle set in the boost process, and each particle in the particle set represents the joint state of the power equipment operation characteristics and the environmental variable characteristics;

[0079] Based on the joint state, the transfer matrix between the source intensity distribution and the particle reconstruction surface is constructed, and the observation guidance term is constructed using the pseudo-inverse of the transfer matrix.

[0080] The diffusion range of the particle reconstruction boost process is suppressed based on the covariance matrix so that the observation guidance term satisfies the perturbation amplitude condition of the transfer matrix.

[0081] The random permutation resampling algorithm is used to retain the high-weight particles in the particle set that meet the perturbation amplitude conditions, and at the same time, Gaussian perturbations are performed on the low-weight particles in the particle set to obtain the updated particle set;

[0082] A particle state estimator is constructed based on the updated particle set. The weighted average state of the updated particle set is calculated by the particle state estimator, and the mutual adaptation deviation between the operation of the power equipment and environmental factors is estimated according to the weighted average state.

[0083] It should be noted that calculating the weighted average state of the updated particle set through the particle state estimator and estimating the mutual adaptation deviation between the power equipment and environmental factors includes: calculating the weighted average state of the particle set according to the weight of the particle set. Estimating the mutual adaptation deviation between the power equipment and environmental factors based on the weighted average state; judging the superposition effect between changes in human activity and changes in the operating state of power equipment when sudden human gatherings occur based on the characteristics of human activity, and capturing the local load deviation caused by dense human population, which specifically includes:

[0084] According to the characteristics of personnel activities, determine whether there is a sudden gathering of people. Based on the sudden gathering of people, establish a joint model, integrate the personnel activity data with the operating status data of the power equipment, and use the joint model to predict the local load deviation that may be caused by dense crowds. Based on this prediction, dynamically adjust the operating strategy of the power equipment and take corresponding load management measures to cope with the load fluctuations caused by sudden gatherings of people.

[0085] The total deviation of the power energy consumption forecast result is obtained by integrating the power equipment jump deviation, mutual adaptation deviation and local load deviation, and the total deviation in the power energy consumption forecast result is corrected using the causal reasoning algorithm.

[0086] It should be noted that the use of causal reasoning algorithms to correct the total deviation in the power consumption forecast results includes:

[0087] Step 1: The power equipment jump deviation, mutual adaptation deviation and local load deviation are integrated to obtain the total deviation of the power energy consumption forecast result;

[0088] Step 2: Based on the total deviation of the power consumption forecast results, a causal graph model is constructed using a causal inference algorithm based on kernel density estimation. The probability distribution of the total deviation is obtained through kernel density estimation in the context of causal inference.

[0089] Step 3: Infer the causal relationship between each pair of variables based on the probability distribution of the total deviation, and analyze the strength and direction of the causal relationship between each pair of variables through regression analysis; analyze the causal path between the variables, and identify the key causal factors through the influence of each variable in the causal path; correct the total deviation in the power consumption forecast results based on the inferred causal relationship and the probability density function of the total deviation.

[0090] S3. Build an early warning and control mechanism based on the power consumption forecast results of the smart park, and realize the power resource management of the smart park through the early warning and control mechanism.

[0091] It's important to note that, based on the analysis of forecast results, an intelligent early warning and control system is constructed to achieve efficient management of the park's power resources. When the predicted load reaches a preset threshold, the system automatically triggers a peak warning or abnormal alarm, and flexibly adjusts equipment operating conditions and load distribution strategies based on actual conditions. Integrating load forecast data, the system implements measures such as optimizing power usage hours, deploying green energy, and shutting down non-essential and inefficient equipment to develop energy-saving and consumption-reduction control plans. The results of these control measures are then incorporated into feedback loops to continuously refine the park's energy management strategies.

[0092] When the predicted load approaches or exceeds a preset threshold, the system automatically triggers an early warning mechanism and adjusts equipment operating status and load distribution based on real-time demand. Control strategies include peak load balancing, shutting down inefficient equipment, and distributed energy scheduling, while prioritizing the use of renewable energy to achieve energy conservation and emission reduction goals.

[0093] It should be noted that the core of this invention's artificial intelligence spatiotemporal prediction and intelligent optimization of power consumption in smart parks lies in the use of artificial intelligence technology to combine power demand forecasting with energy optimization scheduling in both time and space dimensions, aiming to reduce energy waste and promote green development. This is specifically reflected in the following aspects:

[0094] 1. Prediction in the time dimension:

[0095] The long-term prediction achieved by the present invention mainly makes macro-estimates of electricity demand based on long-term trends over quarters, years, and even years. This type of prediction is often used to predict the overall energy demand of the park, providing a basis for annual or quarterly energy planning and budgeting. The time-scale prediction achieved by the present invention focuses on the changes in the park's electricity consumption on a weekly or monthly basis, which helps the park make more accurate short-term energy scheduling decisions. It focuses on periodic changes, such as the difference between weekdays and weekends, and the change in activity density at the beginning and end of the month.

[0096] 2. Prediction in spatial dimension:

[0097] The park-wide power demand forecast implemented by the present invention focuses on the overall power consumption of the park, which is affected by multiple factors such as weather, seasonal changes, external power grid power supply conditions, and overall park activities. This type of forecast is often used for the park's annual energy planning and macro-level optimization. The park implemented by the present invention can usually be divided into multiple functional areas (such as office areas, commercial areas, industrial areas, parking lots, etc.), and the power demand patterns of each area vary significantly. Based on the differences in energy consumption patterns in spatial regions, more accurate energy consumption information positioning analysis is carried out.

[0098] 3. Power consumption early warning mechanism:

[0099] This invention achieves real-time monitoring and prediction of power consumption in parks, buildings, or equipment, identifying potential abnormal energy consumption in advance and issuing timely warning signals. This mechanism aims to optimize energy use, avoid unnecessary waste, and ensure stable system operation.

[0100] The classification is as follows:

[0101] ① Low-level warning (yellow): When the energy consumption forecast results show that the energy consumption level of a certain area or equipment is close to the set warning threshold, although it is still within the controllable range, it is still necessary to strengthen monitoring, make adjustments and optimization work to prevent the trend from further deteriorating.

[0102] ② Medium-level warning (orange): When the energy consumption forecast results show that the energy consumption of a certain area or equipment has increased significantly, exceeded the normal level and shows a continuous growth trend, if it is not adjusted in time, it may lead to further increase in energy consumption, affecting the efficiency and stability of the entire system.

[0103] ③ High-level warning (red): When energy consumption forecasts show that energy consumption in a particular area or device significantly exceeds normal levels, approaching or exceeding the set target, this indicates significant energy waste. Immediate emergency response measures are required to prevent uncontrolled energy consumption and other serious consequences.

[0104] According to another embodiment of the present invention, Figure 2 As shown, a smart park power consumption prediction system based on artificial intelligence is also provided, which includes:

[0105] Sample library construction module 1 is used to collect multi-source heterogeneous energy consumption data of the smart park, pre-process the multi-source heterogeneous energy consumption data, and build a migration sample library based on the pre-processed multi-source heterogeneous energy consumption data;

[0106] Power consumption prediction module 2 is used to build a spatiotemporal energy consumption prediction model that integrates the time and space dimensions, and use the sample data in the migration sample library to train the spatiotemporal energy consumption prediction model. The trained spatiotemporal energy consumption prediction model is used to predict the power consumption of the smart park in the future time period to obtain the power consumption prediction results;

[0107] The power resource management module 3 is used to build an early warning and control mechanism based on the power consumption prediction results of the smart park, and realize the power resource management of the smart park through the early warning and control mechanism.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for predicting power consumption of a smart park based on artificial intelligence, characterized in that: The method includes: Using IoT sensor devices to collect multi-source heterogeneous energy consumption data of the smart park, where the multi-source heterogeneous energy consumption data includes power equipment operation data, environmental variable data, and personnel activity data; An outlier detection algorithm is used to remove abnormal data caused by environmental interference factors in multi-source heterogeneous energy consumption data, and a linear interpolation algorithm is used to fill in missing data in multi-source heterogeneous energy consumption data; Perform normalization and noise reduction on the multi-source heterogeneous energy consumption data in sequence to eliminate data dimension differences and high-frequency noise interference, and obtain pre-processed multi-source heterogeneous energy consumption data; Reconstruct the pre-processed multi-source heterogeneous energy consumption data to form a spatiotemporal data matrix that reflects the power consumption patterns of the smart park. Calculate the power consumption of the smart park samples based on the spatiotemporal data matrix, and integrate the calculation results to generate core feature sample data. A migration sample library is built based on core feature sample data, and cross-level sample migration is triggered in the migration sample library based on predefined power consumption forecast requirements; Based on the improved long short-term memory network and superposition theorem layer network, a spatiotemporal prediction model for park power consumption is constructed; A stratified sampling algorithm is used to divide the sample data in the migration sample library into a training set, a validation set, and a test set. The trained spatiotemporal energy consumption prediction model is used to predict the power consumption of the smart park in the future time period, and the power consumption prediction results are obtained. Determine whether there is a prediction bias in the power consumption forecast result. If there is no prediction bias, the power consumption forecast result output by the energy consumption spatiotemporal forecast model is used as the final result. Otherwise, the bias in the power consumption forecast result is corrected based on the potential superposition effect between sample features to obtain a corrected power consumption forecast result, specifically including: Extracting sample features after classification of core feature sample data, wherein the sample features include power equipment operation features, environmental variable features, and personnel activity features; Aggregate the operating status of adjacent power equipment based on the operating characteristics of the power equipment, and analyze the power equipment jump deviation in the start and stop of the power equipment in the aggregation of the operating status of adjacent power equipment; An improved particle filter algorithm is used to simulate the superposition effect between the operating characteristics of power equipment and environmental variable characteristics, and to capture the mutual adaptation deviation between the operation of power equipment and environmental factors; When judging sudden gatherings of people based on their activity characteristics, the superposition effect between changes in people's activities and changes in the operating status of power equipment is used to capture local load deviations caused by dense crowds. The total deviation of the power consumption forecast result is obtained by integrating the power equipment jump deviation, mutual adaptation deviation and local load deviation, and the total deviation in the power consumption forecast result is corrected using the causal reasoning algorithm; Based on the power consumption prediction results of the smart park, an early warning and control mechanism is established to achieve power resource management of the smart park through the early warning and control mechanism.

2. The method for predicting power consumption of a smart park based on artificial intelligence according to claim 1 is characterized in that: The method of constructing a migration sample library based on core feature sample data and triggering cross-level sample migration in the migration sample library according to predefined power consumption forecast requirements includes: Based on the core feature sample data, an initial sample library is constructed and the core feature sample data is classified and stored. According to the classified sample categories, the initial sample library is divided into several sub-libraries according to the time dimension and the space dimension; Build a hierarchical migration tree in each sub-database and calculate the migration offset between the current hierarchical migration tree and each of the remaining hierarchical migration trees; Based on the sample migration offset, the samples covered by the rule feature with the highest confidence are migrated from the current sub-library to the adjacent sub-library. After the migration is completed, the new rule features of the remaining samples are calculated and used as the basis for a new round of iterative migration. This process stops when the number of remaining samples is empty or less than the set threshold. Before the iterative migration of samples, a hierarchical migration tree classifier is constructed, and the priority order of sample migration is formulated according to the predefined power consumption prediction requirements. The hierarchical migration tree classifier triggers the migration of samples in the sub-library according to the priority order.

3. The method for predicting power consumption of a smart park based on artificial intelligence according to claim 2 is characterized in that: Calculating the migration offset between the current level migration tree and each of the remaining level migration trees includes: Extract the target sample set from the core feature sample data based on the spatiotemporal data matrix, calculate the total score value of each gene sample in the target sample set, and sort the total score value of each gene sample to obtain the gene sequence; Extract a non-target sample set from the core feature sample data, calculate the total score value of each non-gene sample in the non-target sample set, and sort the total score value of each non-gene sample to obtain a non-gene sequence; The association analysis algorithm is used to calculate the sorting difference between gene sequences and non-gene sequences in the gene dimension, and the sorting difference is used as the migration offset between the current level migration tree and each of the remaining level migration trees.

4. The method for predicting power consumption of a smart park based on artificial intelligence according to claim 3 is characterized in that: The improved particle filter algorithm is used to simulate the superposition effect between the operation characteristics of power equipment and the characteristics of environmental variables, and to capture the mutual adaptation deviation between the operation of power equipment and environmental factors, including: Randomly generate several particles based on the operating characteristics of power equipment and environmental variables and assign initial weights. Construct a state transition equation to analyze the current particle state quantity, and predict the particle state quantity at the next time point based on a first-order Markov process. Compare the difference between the particle state at the next time point and the current particle state with a preset threshold, update the particle weight based on the comparison result, and calculate the adaptation relationship between the power equipment operation characteristics and the environmental variable characteristics based on the updated particle weight; Particle boosting is reconstructed based on the adaptation relationship, and a particle state estimator is constructed using an improved particle filter algorithm. The source intensity distribution during the particle reconstruction boosting process is identified through the particle state estimator, and the mutual adaptation deviation between the operation of power equipment and environmental factors is identified based on the particle source intensity distribution.

5. The method for predicting power consumption of a smart park based on artificial intelligence according to claim 4 is characterized in that: The particle boosting process is reconstructed based on the adaptation relationship, a particle state estimator is constructed using an improved particle filter algorithm, the source intensity distribution during the particle reconstruction boosting process is identified by the particle state estimator, and the mutual adaptation deviation between the operation of the power equipment and the environmental factors is identified based on the particle source intensity distribution. The method includes: The fusion particles reconstruct the particle set in the boost process, and each particle in the particle set represents the joint state of the power equipment operation characteristics and the environmental variable characteristics; Based on the joint state, the transfer matrix between the source intensity distribution and the particle reconstruction surface is constructed, and the observation guidance term is constructed using the pseudo-inverse of the transfer matrix. The diffusion range of the particle reconstruction boost process is suppressed based on the covariance matrix, so that the observation guidance term satisfies the perturbation amplitude condition of the transfer matrix; The random permutation resampling algorithm is used to retain the high-weight particles in the particle set that meet the perturbation amplitude conditions, and at the same time, Gaussian perturbations are performed on the low-weight particles in the particle set to obtain the updated particle set; A particle state estimator is constructed based on the updated particle set. The weighted average state of the updated particle set is calculated by the particle state estimator, and the mutual adaptation deviation between the operation of the power equipment and environmental factors is estimated according to the weighted average state.

6. The method for predicting power consumption of a smart park based on artificial intelligence according to claim 5 is characterized in that: The time series characteristics of power consumption are extracted through the improved long short-term memory network to capture the periodic change trend of power consumption; the interpretable energy consumption activation function of the theorem layer network is superimposed to reduce the number of parameters in the spatiotemporal prediction model of energy consumption.

7. An artificial intelligence-based smart park power consumption prediction system, using the artificial intelligence-based smart park power consumption prediction method according to any one of claims 1 to 6, characterized in that: The system includes: The sample library construction module is used to collect and preprocess the multi-source heterogeneous energy consumption data of the smart park, and build a migration sample library based on the preprocessed multi-source heterogeneous energy consumption data; The power consumption prediction module is used to build a spatiotemporal energy consumption prediction model that integrates the time and space dimensions. It uses the sample data in the migration sample library to train the spatiotemporal energy consumption prediction model. The trained spatiotemporal energy consumption prediction model is used to predict the power consumption of the smart park in the future time period to obtain the power consumption prediction results. The power resource management module is used to build an early warning and control mechanism based on the power consumption prediction results of the smart park, and realize the power resource management of the smart park through the early warning and control mechanism.

Images