Energy consumption optimization method and system based on multi-dimensional data fusion

Abstract

This invention relates to the technical field of smart energy management, and discloses an energy consumption optimization method and system based on multi-dimensional data fusion. The method includes: real-time collection of energy consumption data, environmental data, and equipment operating status data of a target area, followed by data preprocessing to generate a sample dataset; integrating the sample dataset into an energy consumption feature matrix using a multi-dimensional data fusion strategy combining feature-level fusion and decision-level fusion based on spatiotemporal correlation analysis; inputting the matrix into an energy consumption prediction model constructed based on a deep learning algorithm, and outputting energy consumption prediction results for the target area in future periods; and generating an optimal energy consumption control strategy based on the energy consumption prediction results using a dynamic programming algorithm. This invention generates a dynamic optimal energy consumption control strategy through the linkage of deep learning and dynamic programming algorithms, solving the problems of insufficient accuracy and poor adaptability in existing energy consumption optimization technologies.

Description

Technical Field

[0001] This invention relates to the technical field of smart energy management, and in particular to an energy consumption optimization method and system based on multi-dimensional data fusion. Background Technology

[0002] Currently, traditional energy management systems typically employ a single type of data source and fixed parameter control strategies, relying solely on historical experience or simple statistical methods for prediction and control. This approach is ill-suited to the highly dynamic and complex nature of energy consumption scenarios.

[0003] In recent years, some studies have attempted to use machine learning or data-driven methods for energy consumption prediction and optimization, but most of them are still limited to single-dimensional data or simple fusion strategies, lacking in-depth exploration of the complex correlation characteristics between different dimensions of data, resulting in low prediction accuracy and insufficient generalization ability of control strategies.

[0004] Existing energy consumption optimization technologies mainly suffer from the following problems: First, the data collection dimension is singular, only considering historical energy consumption data and ignoring the direct impact of environmental factors and equipment operating status on energy consumption, making it difficult to accurately characterize the actual changes in energy consumption; second, there is a lack of effective data fusion mechanisms, with existing technologies mostly employing simple data splicing or single-level fusion methods, failing to fully explore the intrinsic correlations between different types of data in the spatiotemporal dimension, resulting in insufficient feature extraction and inadequate prediction accuracy; third, current control strategies are usually based on static rules or single algorithm designs, which are slow to respond to real-time changes in energy consumption demand and abnormal situations, unable to achieve dynamic parameter adjustment and energy allocation optimization, and even more difficult to quickly identify and warn of abnormal energy consumption, making it difficult to achieve optimal energy-saving effects.

[0005] Therefore, how to effectively integrate multi-source heterogeneous data and combine advanced prediction models and intelligent optimization algorithms to form a precise, efficient, and dynamic energy consumption optimization scheme has become an urgent technical problem to be solved.

[0006] In summary, existing energy management technologies based on single-dimensional data or simple data fusion methods suffer from low accuracy in energy consumption prediction and a lack of dynamic adaptability and timeliness in energy consumption control strategies. The energy consumption optimization method based on multi-dimensional data fusion provided by this invention introduces a real-time multi-dimensional data acquisition mechanism from the Internet of Things and combines feature-level fusion and decision-level fusion strategies to perform deep correlation analysis on energy consumption, achieving high-precision energy consumption prediction. Furthermore, it generates a dynamic optimal energy consumption control strategy through the linkage of deep learning algorithms and dynamic programming algorithms, thus solving the problems of insufficient accuracy and poor adaptability in energy consumption optimization in existing technologies. Summary of the Invention

[0007] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0008] In view of the aforementioned existing problems, the present invention is proposed.

[0009] To solve the above-mentioned technical problems, the present invention provides the following technical solution: real-time collection of energy consumption data, environmental data and equipment operating status data of the target area through IoT terminal devices, and data preprocessing to generate a sample dataset;

[0010] Based on spatiotemporal correlation analysis, a multi-dimensional data fusion strategy combining feature-level fusion and decision-level fusion is adopted to integrate the sample dataset into an energy consumption feature matrix;

[0011] Input the energy consumption feature matrix into the energy consumption prediction model built based on deep learning algorithm, and output the energy consumption prediction results for the target area in future time periods;

[0012] Based on the energy consumption prediction results, an optimal energy consumption control strategy is generated through a dynamic programming algorithm. The optimal energy consumption control strategy includes equipment operating parameter adjustment, energy allocation optimization, and abnormal energy consumption alarm.

[0013] As a preferred embodiment of the energy consumption optimization method based on multi-dimensional data fusion described in this invention, the energy consumption data includes electricity consumption, water consumption, gas consumption, and energy consumption data obtained from other energy consumption monitoring instruments.

[0014] The environmental data includes indoor and outdoor temperature, humidity, air quality index, illuminance intensity, and meteorological conditions around the target area;

[0015] The equipment operating status data includes equipment start / stop status, operating mode, power load level, fault information, and maintenance cycle information.

[0016] As a preferred embodiment of the energy consumption optimization method based on multi-dimensional data fusion described in this invention, the collected energy consumption data, environmental data, and equipment operating status data are preprocessed to generate a sample dataset, including:

[0017] Missing values ​​are filled in using the same feature interpolation method;

[0018] If the detected outliers exceed the upper and lower limits obtained based on statistical analysis, they are smoothly replaced according to the historical average within the context period.

[0019] Kalman filtering is used to smooth the data in order to eliminate the interference of random noise;

[0020] Normalization methods are used to transform data of different dimensions into a uniform numerical range;

[0021] Based on the recorded timestamps, the collected multi-source data is aligned, and the consistency is determined by comparing the sum of the sensor master table and the sub-tables. If a conflict occurs, the conflicting data is removed.

[0022] The preprocessed data is arranged in ascending order of time to form a multi-source sample dataset.

[0023] As a preferred embodiment of the energy consumption optimization method based on multidimensional data fusion described in this invention, based on spatiotemporal correlation analysis, a multidimensional data fusion strategy combining feature-level fusion and decision-level fusion is adopted to integrate the sample dataset into an energy consumption feature matrix, including:

[0024] For the energy consumption changes of the same device and the same measuring point in a continuous time series, the energy consumption data and the environmental data are combined with the lag window to extract time series features, thereby obtaining the time series features of energy consumption changes with time and with environmental conditions.

[0025] Correlation analysis is performed on the operating status data of each device within the target area to obtain the local energy consumption distribution characteristics;

[0026] The temporal features and local energy consumption distribution features at the same time are weighted and aggregated to obtain a preliminary feature combination;

[0027] The energy consumption state determination results corresponding to different feature combinations are weighted and averaged to obtain the optimal energy consumption feature matrix.

[0028] As a preferred embodiment of the energy consumption optimization method based on multidimensional data fusion described in this invention, the multidimensional data fusion strategy includes:

[0029] Principal component analysis is used to project and compress the high-dimensional feature space to remove redundant features.

[0030] Based on the differences in energy consumption and environment at various locations within the target area, a spatial distance attenuation coefficient is set, and the characteristics of adjacent spatial units are weighted in a hierarchical manner.

[0031] Data from multiple sensors within the same time period are fused by assigning different weights based on reliability to obtain a more stable feature representation.

[0032] The energy consumption feature matrix generated after fusion is checked for consistency and integrity. If feature conflicts are found, the process is repeated to correct the conflicting features.

[0033] As a preferred embodiment of the energy consumption optimization method based on multi-dimensional data fusion described in this invention, the energy consumption feature matrix is ​​input into an energy consumption prediction model constructed based on a deep learning algorithm, and the energy consumption prediction results for the target area in future time periods are output, including:

[0034] The energy consumption feature matrix is ​​divided into a training set, a validation set, and a test set based on time series and device type;

[0035] The energy consumption feature matrix is ​​successively input into the deep learning model;

[0036] Iterative training is performed using the stochastic gradient descent algorithm, and backpropagation is performed using the error between the actual value of historical energy consumption data and the model output value.

[0037] If overfitting occurs due to the reduction in validation set error during each training iteration, the model parameters will be automatically adjusted according to the set learning rate decay strategy.

[0038] Using a pre-trained deep learning model, the energy consumption of the target area in the short, medium and long term is predicted, and the prediction results are output.

[0039] As a preferred embodiment of the energy consumption optimization method based on multidimensional data fusion described in this invention, the prediction results include an overall energy consumption prediction curve and sub-item energy consumption prediction values ​​and confidence intervals divided according to different equipment types.

[0040] As a preferred embodiment of the energy consumption optimization method based on multi-dimensional data fusion described in this invention, an optimal energy consumption control strategy is generated using a dynamic programming algorithm based on the energy consumption prediction results. The optimal energy consumption control strategy includes equipment operating parameter adjustment, energy allocation optimization, and abnormal energy consumption alarm; wherein:

[0041] The real-time monitored actual energy consumption value E t The predicted value output by the energy consumption prediction model Perform a comparison; if the conditions are met... This is then determined to be abnormal energy consumption, where σ h β represents the historical standard deviation of the prediction error, and β is an adjustable sensitivity factor.

[0042] When abnormal energy consumption occurs, an alarm is immediately triggered and the alarm information is recorded. If the abnormality continues for more than the preset time, the system will automatically switch to emergency handling mode.

[0043] For adjustable equipment, adjustment strategies are generated based on predicted peak and off-peak energy consumption periods and usage demands. These strategies include: increasing / decreasing the set temperature and humidity values ​​of the air conditioner during non-critical periods to reduce cooling / heating energy consumption; shifting some high-energy-consuming operations from peak periods to off-peak periods; and adaptively determining when to enter energy-saving mode, standby mode, or complete shutdown based on the comparison between measured and predicted values.

[0044] The dynamic programming algorithm is used to find the optimal combination of different energy types and quotas in each time period so that the total cost and emissions are balanced.

[0045] Define state in, To predict energy demand at time t, R t For currently available resources, Ω t This refers to the accumulated costs;

[0046] At each discrete time t, a set of optional actions is set, including adjusting equipment load, switching energy supply schemes, and maintaining the current strategy;

[0047] Based on the principle of minimizing the objective function, the state transitions at different times are traversed to obtain the optimal action sequence that minimizes the overall cost within the time period T.

[0048] The optimal action sequence is mapped to executable device control parameters and energy allocation ratios;

[0049] Real-time monitoring of energy consumption changes after execution; when actual energy consumption deviates too much from the predicted value, automatic re-prediction and dynamic programming solution are performed, and the optimal energy consumption control strategy is updated in a timely manner.

[0050] As a preferred embodiment of the energy consumption optimization system based on multi-dimensional data fusion described in this invention, it includes: one or more processors;

[0051] The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the energy consumption optimization method based on multidimensional data fusion as described above.

[0052] As a preferred embodiment of the computer-readable medium for the storage software described in this invention, the software includes instructions executable by one or more computers, the instructions causing the one or more computers to perform operations, the operations including the flow of the energy consumption optimization method based on multidimensional data fusion as described above.

[0053] The beneficial effects of this invention are:

[0054] 1. Through real-time and comprehensive data collection, the actual changes in energy consumption can be reflected in a timely manner. After preprocessing, noise, outliers and missing values ​​in the data are effectively removed, which improves the effectiveness and accuracy of data analysis and lays a solid foundation for subsequent multi-dimensional data fusion, thereby ensuring the accuracy of subsequent prediction and control.

[0055] 2. Through a multi-dimensional fusion strategy, in-depth mining and correlation analysis between data from different dimensions were achieved, fully revealing the complex correlation characteristics between energy consumption data and environmental factors and equipment operating status data. Feature-level fusion effectively reduced data dimensionality, improved the representativeness of features, and avoided interference from redundant data; while decision-level fusion further strengthened the robustness of data fusion. By weighting and deciding on multiple analysis results, the reliability and accuracy of feature fusion were greatly improved.

[0056] 3. Through the prediction step, the powerful nonlinear fitting and generalization capabilities of deep learning algorithms are utilized to capture and predict the dynamic trend of future energy consumption in the target area, achieving high-precision prediction of future energy consumption and avoiding the problems of energy waste or insufficient energy supply caused by insufficient prediction accuracy in traditional methods.

[0057] 4. Through the optimization of dynamic programming algorithm, dynamic planning and optimal control of energy consumption in future time periods are realized. By dynamically adjusting the equipment operating parameters, the operating efficiency of the equipment is optimized in real time, and the operating costs caused by peak load are reduced. Through the energy allocation optimization method, the cost and environmental impact of multiple energy sources can be comprehensively considered to achieve efficient energy utilization. Through abnormal energy consumption alarm, abnormal energy consumption conditions can be quickly identified and responded to, avoiding the spread of abnormal energy consumption or causing more serious resource waste. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the 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. Wherein:

[0059] Figure 1 This is a flowchart illustrating the energy consumption optimization method based on multidimensional data fusion as shown in this invention. Detailed Implementation

[0060] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0061] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0062] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0063] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates an energy consumption optimization method based on multi-dimensional data fusion, which specifically includes the following steps:

[0064] S1. Real-time collection of energy consumption data, environmental data, and equipment operating status data of the target area via IoT terminal devices, followed by data preprocessing to generate a sample dataset. Note the following in this step:

[0065] By deploying IoT terminal devices within the target area, multi-source heterogeneous data is acquired in real time according to a set sampling period (e.g., 10 minutes), including:

[0066] Energy consumption data: electricity consumption, water consumption, gas consumption, and energy consumption data obtained from other energy consumption monitoring instruments;

[0067] Environmental data: indoor and outdoor temperature, humidity, air quality index, illuminance intensity, and surrounding meteorological conditions of the target area;

[0068] Equipment operating status data: equipment start / stop status, operating mode, power load level, fault information, and maintenance cycle information;

[0069] To ensure the timeliness and integrity of the data, the sampling time of each sensor is synchronized with a unified clock, and the data is uploaded to the local data center immediately after each sampling.

[0070] In an optional implementation, missing values ​​caused by sensor malfunctions or network fluctuations during the data acquisition process are filled using an interpolation method with similar characteristics to the current missing values. Based on historical statistical analysis, upper and lower limit ranges for various data are pre-set, and the collected energy consumption data, environmental data, and equipment operating status data are subjected to range checks. If the detected value exceeds the upper or lower limit range, it is considered an outlier. The outlier is then smoothed and replaced with the historical average within the context time period to make it more reasonable to connect with the data of adjacent time periods and avoid interference from extreme data points. The energy consumption data is further smoothed using a Kalman filter algorithm. During the filtering process, the optimal state is dynamically estimated based on the system state transition equation and observation equation to eliminate fluctuations caused by random interference.

[0071] Furthermore, since energy consumption data, environmental data, and equipment operating status data have different numerical dimensions, a normalization method is used to map them to the same numerical range (e.g., [0,1]) to facilitate subsequent fusion and modeling. Energy consumption data, environmental data, and equipment operating status data are aligned according to timestamps to ensure accurate correspondence of multi-source information at the same point in time. The total and sub-tables of sensor measurements are compared, and if conflicting data with significantly inconsistent total measurements occur at the same timestamp, the conflicting records are removed.

[0072] The preprocessed data are arranged in ascending chronological order to form a sample dataset; let T be the total number of sampling times, and let x... t Let represent the feature vector obtained after complete preprocessing at time t. Then, the sample dataset DS is represented as:

[0073] DS = {x t |t=1,2,…,T}

[0074] x t =[E t ,Env t ,S t ]

[0075] Among them, E t Env represents the energy consumption data vector collected at time t. t S represents the environmental data vector collected at time t. t This represents the vector of device operating status data collected at time t;

[0076] After normalization and temporal alignment, the above sub-vectors constitute a comprehensive feature description under the same timestamp.

[0077] S2. Based on spatiotemporal correlation analysis, a multi-dimensional data fusion strategy combining feature-level fusion and decision-level fusion is adopted to integrate the sample dataset into an energy consumption feature matrix. Note that the following should be noted in this step:

[0078] The energy consumption changes of the same equipment and the same measuring points in the sample dataset over a continuous time series are analyzed. A lag window w is introduced between the energy consumption data and the environmental data. t This is used to capture the dynamic characteristics of energy consumption changing over time and with varying environmental conditions, forming a time-series feature vector whose elements include the time from the previous moment to the past moment w. t Energy consumption and key environmental indicators within a given time period are represented by time-series feature vectors through statistical analysis, correlation analysis, and smoothing to obtain their feature representations in the time dimension.

[0079] Perform correlation analysis on the operating status data of various devices within the target area at the same time scale, and assume... This represents the operating status data of the i-th device at time t. Devices with high interrelationships are grouped and analyzed to form local energy consumption distribution characteristics, which include the overall load of the local device group, fault concentration, and maintenance cycle urgency.

[0080] The extracted local energy consumption distribution features and time-series feature vectors are weighted and aggregated at the same time t to obtain a preliminary feature combination;

[0081] In an optional implementation, a multi-dimensional data fusion strategy is used to further process the initially obtained feature combinations, specifically including:

[0082] Principal component analysis is performed on the high-dimensional feature space to remove redundant features, resulting in projection results with relatively lower dimensionality and more concentrated information.

[0083] Considering the differences in energy consumption and environment at different locations within the target area, a distance attenuation coefficient is introduced for adjacent spatial units, and higher weights are assigned to the characteristics of measuring points or equipment that are close to each other.

[0084] If there are differences in the observations of multiple sensors within the same time period, different weights are allocated according to historical reliability to improve the stability of the fusion results.

[0085] Based on the weighted results above, the deep fusion of the initial feature combination is completed. If feature conflict is detected (such as the same measurement point showing obvious contradictory values ​​at the same time), the relevant conflicting features are backtracked and corrected until the consistency requirements are met.

[0086] The energy consumption status of the fused feature combination is determined in multiple dimensions, and the energy consumption level, environmental factors and equipment load at the current time t are classified according to the set threshold.

[0087] If different feature combinations give different judgment results at time t, then the judgment results are processed by weighted average, and the results with higher spatial correlation are given greater weight to obtain the optimal time judgment vector.

[0088] By integrating temporal feature information, local distribution features, and the optimal judgment result, a unified time fusion vector h is generated. t ;

[0089] The fused vector h at each time point t Arrange them in chronological order (t = 1, 2, ..., T) to obtain the final energy consumption characteristic matrix H;

[0090] If h t ∈R D Let the fused D-dimensional feature vector be represented, then the energy consumption feature matrix H is expressed as:

[0091]

[0092] Where T is the total number of sampling times, and D is the final dimension of the fused feature vector;

[0093] It should be noted that the energy consumption feature matrix H contains comprehensive information on spatiotemporal correlation, device correlation, and multi-source environmental fusion, providing a high-precision and highly consistent data foundation for subsequent input into deep learning models and dynamic programming optimization.

[0094] S3. Input the energy consumption feature matrix into the energy consumption prediction model built based on the deep learning algorithm, and output the energy consumption prediction results for the target area in the future. Note that the following points should be noted in this step:

[0095] The energy consumption feature matrix H obtained from step S2 is divided into a training set, a validation set, and a test set according to the time series. For example, if 70% of the data is used for training, 15% for validation, and 15% for testing, and the total time length of H is T, and each record contains D-dimensional features, then the training set contains 0.7T continuous time series data, and the validation and test sets each contain 0.15T data.

[0096] Define an energy consumption prediction model M based on a Long Short-Term Memory (LSTM) network, which consists of an input layer, an LSTM hidden layer, and an output layer;

[0097] For the feature vector h input at time t t ∈R D The LSTM unit updates its hidden state H t With memory unit C t And mapped to predicted values ​​in the output layer.

[0098] Energy consumption prediction model M, at time t, is based on the input feature h t Output predicted energy consumption value Further combinations are used to form multi-period forecasts;

[0099] Assume the training set contains N time series records, each record corresponding to a time point h. t Compared with the historical energy consumption value y t , where h t ∈R D Let y represent the D-dimensional eigenvector at time t. t ∈R K This represents the actual values ​​of the corresponding K energy consumption sub-items;

[0100] Let the initial learning rate be η0, for example, η0 = 0.001. Set the batch size to 32, that is, select 32 training data in each batch for one forward calculation and one backward update.

[0101] The maximum number of training epochs is set to 100. Before the first epoch, the LSTM model and output layer parameters (such as the weight vector W) are adjusted. o Bias vector b o Perform random, small-scale initialization;

[0102] During each training iteration, a batch of data {h} is taken from the training set. t ,y t}, all h in this batch t Input the data into the LSTM cell in the correct timing sequence:

[0103] (H t C t ) = LSTM(h t H t-1 C t-1 ;θ)

[0104] Among them, H t Let C be the hidden layer state vector at the current time. t Let θ be the state vector of the memory unit, and let θ represent the trainable parameters of the LSTM (such as the weights and biases of the forget gate, input gate, and output gate).

[0105] The LSTM output layer will store the hidden layer state H t Mapped to predicted values

[0106]

[0107] in, b o ∈R K ;

[0108] Compare the predicted values ​​with the actual values ​​at all times t within the batch, and calculate the loss function L(θ). When using mean squared error (MSE), it is denoted as:

[0109]

[0110] Where ||·|| denotes the vector 2 norm, y t and These are the actual value and the predicted value, respectively.

[0111] Using the stochastic gradient descent algorithm to analyze θ and W o b o Perform backpropagation updates:

[0112]

[0113] Where η is the current learning rate;

[0114] After completing one batch, continue to the next batch until all training samples have been traversed;

[0115] During training, if the validation set error starts to rise significantly or shows no decreasing trend, overfitting may occur. In this case, the learning rate decay strategy is activated, updating the current learning rate η to... in (like );

[0116] If the degradation continues and cannot be improved, an early stopping strategy is triggered to terminate training and retain the parameters of the model with the best performance.

[0117] For example, the initial learning rate η0 = 0.001, the batch size is 32, the maximum number of iterations is 100, and the decay coefficient γ = 0.9;

[0118] At the end of the first round of training, the training set MSE (mean squared error) was 25.3, and the validation set MSE was 28.7.

[0119] By the 5th round, the validation set MSE dropped to 15.2. After continuing training to the 10th round, the validation set MSE approached 12.9, at which point oscillations began to appear.

[0120] Learning rate decay was triggered in the 12th round, η was updated to 0.0009, and training continued until the 20th round, with the validation set MSE stabilizing at 9.8;

[0121] If the validation set MSE no longer decreases significantly after round 22, training is stopped early, and the model parameters at this point are saved, resulting in the final trained LSTM energy prediction model M. * .

[0122] It should be further noted that the output includes energy consumption forecasts for the target area in future periods, including short-term, medium-term, and long-term forecasts, among which:

[0123] Short-term forecasting uses M * Predicting energy consumption for the next hour, Indicates short-term forecast results;

[0124] Assume the prediction result for the next hour is as follows:

[0125]

[0126] These figures represent energy consumption of 28.5 kWh for air conditioning, 6.3 kWh for lighting, and 3.2 kWh for production equipment, respectively.

[0127] If confidence interval If estimated at ±1kWh, the short-term forecast range is expressed as [27.5, 29.5]kWh (air conditioning portion), and so on;

[0128] The medium-term forecast will be h t Given a feature input sequence for the next day (24 hours), the LSTM model is repeatedly called for progressive rolling prediction to obtain...

[0129] If the total energy consumption forecast for one day later is:

[0130]

[0131] Of these, air conditioning accounted for 440 kWh, lighting accounted for 200 kWh, and production equipment accounted for 80 kWh;

[0132] Long-term forecasts, which predict energy consumption trends over several weeks (e.g., one week), are achieved through multiple rolling forecasts, such as predicting the cumulative energy consumption for the following week. Air conditioning accounts for the highest energy consumption.

[0133] Then, the predicted values ​​of each time point t are arranged in chronological order to form the overall energy consumption prediction curve and the sub-item energy consumption prediction curve;

[0134] If let Let represent the predicted value of the i-th energy consumption item at time t, then denoted as:

[0135] and corresponding confidence intervals

[0136] Where K represents the total number of energy consumption sub-items. Let be the prediction confidence radius of this sub-item at time t;

[0137] By With y t Continuous comparisons are performed, and an anomaly alarm is triggered if the error deviates too much.

[0138] S4. Based on the energy consumption prediction results, generate the optimal energy consumption control strategy using a dynamic programming algorithm. The optimal energy consumption control strategy includes adjusting equipment operating parameters, optimizing energy allocation, and issuing alarms for abnormal energy consumption. It should be noted that the following points are important in this step:

[0139] set up E is the predicted energy consumption value output from step S3 at time t. t This refers to the actual energy consumption collected in real time.

[0140] If satisfied Then it is determined that abnormal energy consumption has occurred at the current moment;

[0141] Where, σ h The historical standard deviation of the prediction error is represented by β, which is an adjustable sensitivity factor (e.g., 2.0). When abnormal energy consumption is detected, an alarm signal is immediately triggered and the alarm information is recorded in the log.

[0142] If the abnormal situation lasts for more than a preset threshold (e.g., 15 minutes), the system will automatically switch to emergency handling mode and perform the following operations:

[0143] The upper limit of operating power for high-power equipment will be temporarily lowered.

[0144] Trigger dedicated maintenance personnel to inspect the equipment; if any equipment malfunction information is found, prioritize its troubleshooting and repair.

[0145] As an example, in the scenario of a commercial building, σ h ≈3.0kWh, β=2.0, if the current If the power consumption exceeds 6 kWh, an abnormal alarm will be triggered. If the limit is exceeded for more than 15 minutes, the system will switch to emergency mode, automatically shut down some non-critical equipment, and send an alarm SMS to the management personnel.

[0146] In an optional implementation, the prediction result based on step S3 is used. Identify peak and off-peak energy consumption periods on a daily or weekly basis;

[0147] Let P represent the set of peak periods and V represent the set of trough periods. The entire day is divided into segments based on the prediction curve. For example, peak periods are often concentrated between 10:00-12:00 and 14:00-16:00 on weekdays.

[0148] For adjustable equipment (such as air conditioning, lighting, and ventilation units), adjust operating parameters appropriately during non-critical periods:

[0149] For example, during off-peak hours, appropriately reduce the frequency of air conditioning operation, increase the air conditioning cooling temperature, or reduce the brightness of lighting;

[0150] Some high-energy-consuming operations will be transferred from P to V for execution.

[0151] Based on the measured value E t Compared with the predicted value The system continuously compares data, and if it finds that the device load is significantly higher than the predicted level, it automatically switches to standby mode until the next detection cycle.

[0152] As an example, if the air conditioner is set to 24°C, it can be raised to 26°C in the evening (during low energy consumption periods). If a peak air conditioner load is predicted between 12:00 and 13:00, the temperature can be raised to 25°C at 11:30 to reduce peak electricity consumption.

[0153] Furthermore, when the target area has multiple energy supplies (such as electricity, municipal heating, solar energy, and energy storage devices), it enables... This represents the unit cost of using energy j at time t. This represents the unit emissions when energy j is used at time t;

[0154] Define α and γ as balance coefficients, respectively, to measure the importance of cost and emissions, and define the optimization objective function:

[0155]

[0156] in, This represents the amount of energy used at time t, where J is the total number of available energy types and T is the total number of steps in the planning period.

[0157] Define state in, To predict energy demand at time t, R t For currently available resources (such as available energy storage capacity, maximum solar power), Ω t This refers to the accumulated costs;

[0158] For example, the action set includes:

[0159] Adjust the power level of adjustable load devices;

[0160] Switching energy supply solutions (such as switching from the grid to a combination of solar power and energy storage);

[0161] Maintain the current strategy.

[0162] At each time t, based on the principle of minimizing the objective function, a search is performed on all possible action sequences, and a backtracking approach is used to find the optimal sequence that minimizes the overall cost from time t=1 to t=T.

[0163] The final optimal decision includes the energy usage at each time point. And corresponding equipment load adjustment schemes.

[0164] As an example, if the system has three energy options at the same time: grid, solar energy (daytime power limit of 10kW), and energy storage (remaining capacity of 20kWh), then during the midday hours when the sun is strong, solar energy will be used first, supplemented by energy storage, to reduce dependence on grid power; when the electricity price is low at night, grid power will be used to supplement energy storage.

[0165] When α = 1 and γ = 5, it means that the emission weight is significantly higher than the cost, making the system tend to prioritize the use of clean energy; when α = 2 and γ = 1, it focuses more on minimizing economic costs.

[0166] The optimal action sequence obtained by the dynamic programming algorithm The mapping to the optimal energy consumption control strategy specifically includes:

[0167] The scheduling of equipment operating parameters over time;

[0168] Energy allocation plan;

[0169] Abnormal alarm threshold;

[0170] If long-term planning (such as one week) is required, the optimal action sequence for each day will be summarized to generate a schedule for the system to execute on a daily basis.

[0171] After the control policy is issued, the system continuously monitors. With E t The differences in energy consumption and the usage status of each energy source will be considered. If the real-time energy consumption deviates significantly from the predicted value, the system will automatically re-predict and re-execute dynamic programming to adjust the strategy in a timely manner.

[0172] During execution, if a new fault alarm or extreme weather condition is detected, the emergency mode will be triggered and the optimal strategy will be recalculated immediately within the feasible range to ensure the flexibility of energy consumption control.

[0173] For example, in an enterprise factory scenario, the optimal energy consumption control strategy is stored as a table of time period, equipment parameters, and energy sources, as shown in the table below:

[0174] Table 1. Time Period - Equipment Parameters - Energy Source Table

[0175]

[0176]

[0177] Simultaneously record the energy consumption error range for the purpose of determining abnormal alarm thresholds;

[0178] Define a time period set t = 1, ..., T, where dynamic programming provides the device scheduling vector for each time period. and energy allocation vector

[0179] The decision combination for all time periods is represented as:

[0180] Π * ={(u t ,x t ),t=1,…,T}

[0181] Where E is the number of adjustable devices, J is the number of selectable energy types, and u t Includes air conditioning temperature settings, lighting power settings, and equipment start / stop mode information. t This corresponds to the amount of each energy source used in time period t;

[0182] The optimal energy consumption control strategy Π * It has comprehensive optimization results for peak and valley energy consumption, equipment parameters, abnormal alarm thresholds, and multi-energy balance.

[0183] Preferably, the method of the present invention, through the implementation of the above steps, solves the problems of insufficient energy consumption prediction accuracy, single control strategy and delayed response in the prior art, and realizes the effective synergy of comprehensive data sources, deep data fusion, accurate prediction model and intelligent control strategy, thereby achieving significant beneficial effects of improving energy utilization efficiency, reducing energy consumption costs and real-time accurate optimization control.

[0184] The data preprocessing and feature extraction methods for the aforementioned collected data can be carried out using existing technologies and methods, and will not be elaborated further in this example.

[0185] In addition to the above embodiments, other aspects of the present invention also propose an energy consumption optimization system based on multi-dimensional data fusion, comprising: one or more processors and a memory.

[0186] The memory is used to store operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the energy consumption optimization method based on multidimensional data fusion described in the foregoing embodiments, especially... Figure 1 The flowchart of the method is shown.

[0187] Other aspects disclosed in the embodiments of the present invention also propose a computer-readable medium for storing software including instructions executable by one or more computers, which, upon execution, cause the one or more computers to perform operations including the flow of the energy consumption optimization method based on multidimensional data fusion of the foregoing embodiments, particularly... Figure 1The flowchart of the method is shown.

[0188] It should be recognized that embodiments of the present invention may be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium.

[0189] The method can be implemented using standard programming techniques, including a non-transitory computer-readable storage medium configured with a computer program in the computer program, wherein the storage medium is configured such that the computer operates in a specific and predefined manner.

[0190] Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system; however, if necessary, the program can be implemented in assembly or machine language.

[0191] In any case, the language can be either compiled or interpreted.

[0192] Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit.

[0193] The processes described herein (or variations and / or combinations thereof) can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0194] Furthermore, the method can be implemented in any suitable computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices.

[0195] Various aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether portable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein.

[0196] Furthermore, machine-readable code, or parts thereof, can be transmitted via wired or wireless networks.

[0197] When such media includes instructions or programs that combine with a microprocessor or other data processor to implement the steps described above, the invention described herein includes these and other different types of non-transitory computer-readable storage media.

[0198] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended 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 energy consumption optimization method based on multi-dimensional data fusion, characterized in that, include: The system collects energy consumption data, environmental data, and equipment operating status data of the target area in real time through IoT terminal devices, and preprocesses the data to generate a sample dataset. Missing values ​​are filled in using the same feature interpolation method; for detected outliers, if they exceed the upper and lower limits obtained based on statistical analysis, they are smoothly replaced according to the historical mean within the context time period; Kalman filtering is used to smooth the data to eliminate the interference of random noise. Normalization methods are used to transform data of different dimensions into a uniform numerical range; Based on the recorded timestamps, the collected multi-source data is aligned, and the consistency is determined by comparing the sum of the sensor master table and the sub-tables. If a conflict occurs, the conflicting data is removed. The preprocessed data is arranged in ascending order of time to form a multi-source sample dataset; Based on spatiotemporal correlation analysis, a multi-dimensional data fusion strategy combining feature-level fusion and decision-level fusion is adopted to integrate the sample dataset into an energy consumption feature matrix. This includes: extracting time-series features of energy consumption changes over continuous time series for the same device and the same measuring point, combined with a lag window, to obtain time-series features of energy consumption changes with time and environmental conditions; performing correlation analysis on the operating status data of each device within the target area to obtain local energy consumption distribution features; weighted aggregation of the time-series features and the local energy consumption distribution features at the same time point to obtain preliminary feature combinations; and weighted averaging of the energy consumption status judgment results corresponding to different feature combinations to obtain the optimal energy consumption feature matrix. The multidimensional data fusion strategy includes: using principal component analysis to project and compress the high-dimensional feature space to remove redundant features; setting a spatial distance attenuation coefficient based on the energy consumption and environmental differences at each location within the target area, and performing hierarchical weighting of the features of adjacent spatial units; fusing multi-sensor data within the same time period with different weights based on reliability to obtain a more stable feature representation; and performing consistency and integrity checks on the energy consumption feature matrix generated after fusion, and if feature conflicts are found, returning to correct the conflicting features. Input the energy consumption feature matrix into the energy consumption prediction model built based on deep learning algorithm, and output the energy consumption prediction results for the target area in future time periods; Based on the energy consumption prediction results, an optimal energy consumption control strategy is generated through a dynamic programming algorithm. The optimal energy consumption control strategy includes equipment operating parameter adjustment, energy allocation optimization, and abnormal energy consumption alarm.

2. The energy consumption optimization method based on multi-dimensional data fusion according to claim 1, characterized in that, The energy consumption data includes electricity consumption, water consumption, gas consumption, and energy consumption data obtained from other energy consumption monitoring instruments; The environmental data includes indoor and outdoor temperature, humidity, air quality index, illuminance intensity, and meteorological conditions around the target area; The equipment operating status data includes equipment start / stop status, operating mode, power load level, fault information, and maintenance cycle information.

3. The energy consumption optimization method based on multi-dimensional data fusion according to claim 1, characterized in that, Input the energy consumption feature matrix into an energy consumption prediction model constructed based on a deep learning algorithm, and output the energy consumption prediction results for the target area in future time periods, including: The energy consumption feature matrix is ​​divided into a training set, a validation set, and a test set based on time series and device type; The energy consumption feature matrix is ​​successively input into the deep learning model; Iterative training is performed using the stochastic gradient descent algorithm, and backpropagation is performed using the error between the actual value of historical energy consumption data and the model output value. If overfitting occurs due to the reduction in validation set error during each training iteration, the model parameters will be automatically adjusted according to the set learning rate decay strategy. Using a pre-trained deep learning model, the energy consumption of the target area in the short, medium and long term is predicted, and the prediction results are output.

4. The energy consumption optimization method based on multi-dimensional data fusion according to claim 3, characterized in that, The prediction results include the overall energy consumption prediction curve and the sub-item energy consumption prediction values ​​and confidence intervals divided by different equipment types.

5. The energy consumption optimization method based on multi-dimensional data fusion according to claim 3, characterized in that, Based on the energy consumption prediction results, an optimal energy consumption control strategy is generated using a dynamic programming algorithm. This optimal energy consumption control strategy includes adjusting equipment operating parameters, optimizing energy allocation, and issuing abnormal energy consumption alarms. Wherein: Real-time monitored actual energy consumption value The predicted value output by the energy consumption prediction model Perform a comparison; if the conditions are met... If it is abnormal, then it is determined to be abnormal energy consumption. The historical standard deviation of the prediction error. It is an adjustable sensitivity factor; When abnormal energy consumption occurs, an alarm is immediately triggered and the alarm information is recorded. If the abnormality continues for more than the preset time, the system will automatically switch to emergency handling mode. For adjustable equipment, adjustment strategies are generated based on predicted peak and off-peak energy consumption periods and usage demands. These strategies include: increasing / decreasing the set temperature and humidity values ​​of the air conditioner during non-critical periods to reduce cooling / heating energy consumption; shifting some high-energy-consuming operations from peak periods to off-peak periods; and adaptively determining when to enter energy-saving mode, standby mode, or complete shutdown based on the comparison between measured and predicted values. The dynamic programming algorithm is used to find the optimal combination of different energy types and quotas in each time period so that the total cost and emissions are balanced. Define state ,in, To predict the energy demand at time t, For currently available resources, This refers to the accumulated costs; At each discrete time t, a set of optional actions is set, including adjusting equipment load, switching energy supply schemes, and maintaining the current strategy; Based on the principle of minimizing the objective function, the state transitions at different times are traversed to obtain the optimal action sequence that minimizes the overall cost within the time period T. The optimal action sequence is mapped to executable device control parameters and energy allocation ratios; Real-time monitoring of energy consumption changes after execution; when actual energy consumption deviates too much from the predicted value, automatic re-prediction and dynamic programming solution are performed, and the optimal energy consumption control strategy is updated in a timely manner.

6. An energy consumption optimization system based on multi-dimensional data fusion, characterized in that, include: One or more processors; The memory stores operable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, including the flow of the energy consumption optimization method based on multidimensional data fusion as described in any one of claims 1 to 5.

7. A computer-readable medium for storing software, characterized in that: The software includes instructions executable by one or more computers, which, through execution, cause the one or more computers to perform operations, including the flow of the energy consumption optimization method based on multidimensional data fusion as described in any one of claims 1 to 5.

Images