Energy efficiency optimization method based on multi-source data fusion

Abstract

The invention discloses an energy efficiency optimization method based on multi-source data fusion, which relates to the technical field of industrial energy efficiency management and comprises the steps of multi-source data acquisition, preprocessing, feature extraction, feature level fusion, decision level fusion, optimization strategy generation and feasibility verification. The multi-source data acquisition module acquires equipment operation data, environment data and energy consumption data; the preprocessing module processes data through cleaning, format unification and time-space alignment; the feature extraction module generates statistics, time sequence and cross correlation features; the feature level fusion adopts a graph attention network to capture feature association, and the decision level fusion is output through a weighted average integration sub-model; the optimization strategy generation module combines the two types of fusion results to generate an environment regulation and control strategy and a rotating speed regulation and control strategy; and the feasibility verification module verifies the strategy feasibility through cost income analysis. According to the method, the energy consumption prediction precision is improved through multi-source data fusion, short-term and long-term benefits are considered, dynamic optimization of industrial production energy efficiency is realized, and energy consumption redundancy is remarkably reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial energy efficiency management, in particular to an energy efficiency optimization method based on multi-source data fusion BACKGROUND

[0002] In the development process of industrial production energy efficiency management, the requirement for accurate energy consumption regulation is increasing; with the expansion of industrial scale and the rise of equipment complexity, the energy consumption data in the production process has exploded, and the traditional energy efficiency management method relying on manual experience has been difficult to meet the demand in terms of accuracy and real-time performance. The energy efficiency optimization technology based on multi-source data fusion has become a key means to realize industrial energy saving and consumption reduction due to its comprehensiveness and intelligence.

[0003] The traditional scheme adopts single device parameter adjustment and fixed period energy management, etc. The single device parameter adjustment records the device running data such as motor speed and temperature through manual inspection, and sets the control threshold according to experience. The fixed period management controls the energy consumption according to the historical energy consumption peak value, and forcibly limits the production or reduces the load in the high consumption period to achieve the purpose of energy consumption control.

[0004] However, the traditional energy efficiency management scheme has many problems: on the one hand, single parameter adjustment does not consider the environmental and equipment coordination, and the optimization effect is limited; on the other hand, the fixed period management lacks flexibility and easily affects production efficiency; and both of them do not establish a data correlation model, which cannot dynamically adapt to the working condition changes, resulting in the contradiction between energy consumption redundancy and production demand being prominent; therefore, it is crucial to develop a dynamic energy efficiency optimization method that integrates multi-source data and considers cost and benefit. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an energy efficiency optimization method based on multi-source data fusion to solve the problems raised in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: an energy efficiency optimization method based on multi-source data fusion, comprising: Multi-source data acquisition, by recording a specified scene as an energy efficiency optimization target scene, obtaining multi-source data in the scene, including: device running sensor data, scene environment data, and energy consumption data; Data preprocessing and feature engineering, by preprocessing the collected multi-source data, obtaining standardized data after cleaning, heterogeneous unification and space-time alignment, and by feature extraction on the standardized data, obtaining feature data; Multi-source data fusion model construction, by fusing the feature data, constructing an intelligent fusion model; Energy efficiency optimization strategy design, by the output data of the fusion model, obtaining an executable energy efficiency optimization strategy; Model validation and feasibility analysis: The proposed method is compared with traditional methods using real-world scenario data and public datasets to ensure its feasibility. Data management and iteration involves real-time storage and management of various types of data throughout the process, summarizing the core innovations of the method, and planning the iterative direction of the method based on feedback from actual applications and technological developments to continuously improve energy efficiency optimization.

[0007] Preferably, the multi-source data specifically includes: The equipment operation sensor data includes: motor speed obtained through the motor's built-in speed sensor, and running time, number of start-stop cycles, and operating temperature collected through the equipment's built-in sensors; Scene environment data includes: ambient temperature obtained by uniformly distributing temperature sensors within the scene, and scene device density obtained by considering scene area and number of devices; Energy consumption data includes electricity consumption obtained through smart meters.

[0008] Preferably, the preprocessing of the collected multi-source data specifically includes: data cleaning, comprising: Motor speed, using The rule identifies anomalies. For missing data, it imputes the missing data by taking the average of the data points adjacent to the missing point. The specific outlier determination is as follows: ,in The mean, Standard deviation; For operating temperature, data exceeding the equipment threshold are considered extreme values ​​and are replaced with the average value of the same model. Missing data is filled with linear interpolation. Specifically:

[0009] Where t is the current missing time point, t-1 is the previous time point, and t+1 is the next time point. For the temperature data at the next time point, This refers to the temperature data from the previous time point; Ambient temperature fluctuations and noise are corrected using a weighted average; sensor failures are filled using a historical Gaussian distribution model. Equipment density is recalculated after deduplication of equipment numbers. The density is the ratio of the number of equipment to the scene area. Power consumption is assessed by detecting abrupt changes using a sliding window and replacing the abrupt change value with the median value within the window. If an entire segment is missing, its theoretical power consumption needs to be calculated, specifically as follows:

[0010] in, Motor power, Runtime; The format standardization includes unifying the timestamp to "yyyy-MM-dd-HH:mm:ss"; converting all data types to floating-point; and integrating the structure into a table of "timestamp + device ID + data type + value". Spatiotemporal alignment includes time alignment, which uses device data frequency as a benchmark and interpolates low-frequency data; power consumption is divided into equal intervals; and spatial alignment, which binds environmental data according to the region where the device is located.

[0011] Feature extraction includes: basic statistical features, time-series features, and cross-correlation features. Specifically, the basic statistical features extracted include: motor speed, specifically the hourly average. Standard deviation Peak duration is the ratio of peak duration to total duration; ambient temperature includes daily average. Temperature change rate Power consumption includes: peak hour percentage, which is the ratio of peak hour power consumption to total power consumption; The extraction of time-series features specifically involves: first-order difference of rotational speed. Trend slope Power consumption per unit speed ; The specific cross-correlation feature is: environmental-working temperature difference. Speed-energy consumption fitting slope x represents rotational speed, and y represents power consumption; A time-consistent, error-free structured dataset is obtained through data standardization. The aforementioned feature data is then normalized to form a multidimensional matrix. .

[0012] Preferably, the intelligent fusion model includes feature-level fusion and decision-level fusion; wherein, feature-level fusion is constructed based on graph attention networks; The specific steps of the decision-level fusion are as follows: Sub-models are established for equipment, environment, and energy consumption data respectively, and the outputs of each sub-model are obtained. Weights are allocated based on the accuracy of the sub-models using the inverse error method. , The final output is the predicted energy consumption value. .

[0013] Preferably, the specific steps of the energy efficiency optimization strategy are as follows: Comparison of energy consumption forecasts With target energy consumption The deviation was obtained. And further, it can be used to judge whether energy consumption is reasonable based on the deviation; If the discrepancy is unreasonable, trace the source of the deviation, combine the sub-model output to deduce the key influencing factors, and calculate the sub-model's contribution to the deviation percentage. ,like If the contribution deviation accounts for more than 50%, the equipment operating parameters should be optimized. If the contribution deviation accounts for more than 30%, environmental regulation should be optimized. If the contribution deviation exceeds 20%, energy dispatch will be optimized. Based on the contribution deviation ratio of the sub-models, corresponding optimization strategies are generated, namely: equipment parameter optimization, environmental control optimization, and energy dispatch optimization. Specifically, equipment parameter optimization involves: based on the correlation model... The target rotational speed within the safety threshold is obtained. And reduce the number of start-stop cycles; environmental control optimization is applicable to the dominant environmental deviation, that is, temperature T is positively correlated with y, then the target temperature must meet the following requirements. And satisfy Within the normal operating range; energy dispatch optimization, applicable to energy consumption time sequence deviation, combined with peak and valley electricity pricing, to transfer high energy-consuming processes to valley hours; Preferably, the model validation and feasibility analysis method specifically involves: calculating the execution cost C based on the adjusted strategy. ,in , and These represent the costs of implementing each optimization strategy, if If so, it means that the feasibility is met.

[0014] Preferably, the data management and iteration specifically involves: constructing an associated index using a unified timestamp and device ID to form a complete chain of "raw data - feature data - model output - optimization strategy", storing it in the database, and making dynamic adjustments in real time.

[0015] Preferably, the dynamic adjustment specifically involves: after executing the optimization strategy, real-time monitoring of actual energy consumption. ,like Then, the correlation model is refitted, and the parameters are iteratively optimized.

[0016] The technical effects and advantages of this invention are as follows: 1. Integrating multi-source data improves the accuracy of energy efficiency optimization. By integrating equipment operation, environmental and energy consumption data, it breaks through the limitations of traditional single data dimensions. With the help of a dual fusion mechanism at the feature level and decision level, it fully explores the implicit correlation between data, making energy consumption prediction more in line with actual working conditions, providing a reliable basis for precise control, and solving the optimization deviation problem caused by the one-sided data in traditional methods.

[0017] 2. Dynamic strategy generation takes into account both energy saving and production needs. The environmental control and equipment parameter adjustment strategies generated based on the fusion results can be flexibly adapted to real-time operating conditions. This avoids the impact of fixed-time period control on production efficiency and overcomes the limitations of single parameter adjustment. It reduces energy consumption redundancy while ensuring production continuity, and achieves synergistic optimization of energy saving and production.

[0018] 3. The feasibility verification mechanism ensures the implementation of the strategy. Through cost-benefit analysis, the strategy is evaluated throughout its entire lifecycle. The economics are verified from both short-term and long-term perspectives, ensuring that the optimization measures are executable and effective in practical applications. This solves the problem that traditional energy-saving solutions are difficult to implement due to neglecting costs, and provides industrial enterprises with a scalable energy efficiency optimization path. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the 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.

[0020] Figure 1 This is a schematic diagram of the module connection of the present invention.

[0021] Figure 2 This is a schematic diagram illustrating the process of implementing the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 As shown, the present invention provides an energy efficiency optimization method based on multi-source data fusion, including a multi-source data acquisition module, a data preprocessing and feature engineering module, a multi-data fusion module, an energy efficiency strategy generation module, a strategy verification module, and a data management and iteration module.

[0024] The multi-source data acquisition module is connected to the data processing and feature engineering module, the data preprocessing and feature engineering module is connected to the multi-data fusion module, the multi-data fusion module is connected to the energy efficiency strategy generation module, the energy efficiency strategy generation module is connected to the strategy verification module, and the strategy verification module is connected to the data management and iteration module.

[0025] Please see Figure 2As shown in the flowchart, the process of energy efficiency optimization based on multi-source data fusion is as follows: First, multi-source data is collected. After preprocessing and feature extraction, data fusion is performed to obtain the predicted energy consumption value. Then, by judging whether there is an energy consumption deviation, an optimization strategy is generated. After feasibility verification, the strategy is executed. The actual energy consumption is monitored and it is judged whether it meets the standard. If it meets the standard, it runs stably and the data is archived and the model is iterated before it ends. If it does not meet the standard, it returns to the adjustment strategy, forming a complete closed loop.

[0026] As a preferred embodiment, this embodiment takes a production workshop as the target scenario. The workshop has 5 motors of the same model (numbered M1-M5) and mainly performs parts processing. Energy efficiency optimization needs to be achieved through this method.

[0027] Multi-source data acquisition involves identifying a specified scenario as the target scenario for energy efficiency optimization and acquiring multi-source data within that scenario, including: device operation sensor data, scenario environment data, and energy consumption data. The specific steps for multi-source data acquisition are as follows: collect data from the equipment operation sensors, obtain the rotational speed (unit: r / min) of M1-M5 through the motor's built-in speed sensor, and collect the running time (unit: h), start-stop count (unit: times), and operating temperature (unit: ℃) of each motor through the equipment's built-in sensors. The acquisition frequency is 1 minute / time.

[0028] Collect scene environmental data. Eight temperature sensors are evenly distributed in the workshop to obtain the ambient temperature (unit: °C). The collection frequency is 5 minutes / time. The workshop area is 500㎡ and the number of devices is 20. Therefore, the scene device density is 20 / 500=0.04 units / ㎡.

[0029] Collect energy consumption data: Obtain the total electricity consumption of the workshop (unit: kWh) through smart meters, with a collection frequency of 15 minutes / time.

[0030] Data preprocessing and feature engineering involves preprocessing the collected multi-source data to obtain standardized data that has been cleaned, heterogeneous unified, and spatiotemporally aligned. Feature data is then obtained by extracting features from the standardized data. The collected multi-source data underwent preprocessing, specifically data cleaning. Taking the M1 motor speed data as an example, some data points for a certain hour are: 1450, 1452, 1448, 2000, 1455. The average value is calculated as follows:

[0031] The standard deviation is further calculated as follows: ,because This is not valid, so it is not considered an outlier. However, the actual value of 2000 deviates significantly from other data, which may be due to sensor jitter. We replace it with the sliding average of the five adjacent data points. The average of the first four data points is (1450+1452+1448+2000) / 4=1587.5, so we replace 2000 with 1587.5.

[0032] The operating temperature data for the M2 motor is missing, specifically at 10:05. The temperature before and after 10:04 is 45℃ and at 10:06 it is 47℃. Therefore, the linear interpolation x(10:05) = 45 + (47 - 45) × (10:05 - 10:04) / (10:06 - 10:04) = 46℃.

[0033] In the collection of ambient temperature data, the data from a certain sensor fluctuated greatly. The data from adjacent sensors were 25℃ and 26℃, with distance weights of 0.6 and 0.4, respectively. The corrected temperature was 25×0.6+26×0.4=25.4℃.

[0034] In the collected power consumption data, the data for a certain 15-minute period was 0, which is obviously abnormal. During this period, the total power of the motor was 100kW, and the running time was 0.25h. The theoretical power consumption is calculated as follows: , to fill in the gaps.

[0035] All preprocessed data timestamps are unified to the format "yyyy-MM-dd-HH:mm:ss" and further spatiotemporal alignment is performed, including time alignment and spatial alignment.

[0036] Time alignment is as follows: Ambient temperature is measured every 5 minutes, and data is generated every 1 minute through linear interpolation. For example, if 5:00 is 25℃ and 5:05 is 26℃, then the temperature at 5:01 = 25 + (26 - 25) × (5:01 - 5:00) / (5:05 - 5:00) = 25.2℃. Power consumption is measured every 15 minutes. The total power consumption within a certain 15-minute period is 45 kWh, which is evenly distributed to each 1-minute node, i.e., 45 / 15 = 3 kWh / minute.

[0037] Spatial alignment specifically involves marking motors M1-M5 as area A, and binding the ambient temperature, equipment density, and motor data of this area.

[0038] Furthermore, feature data is extracted, including: basic statistical features, time-series features, and cross-correlation features; Extracting basic statistical features, specifically: motor speed including: hourly average. Standard deviation Peak duration is the ratio of peak duration to total duration; ambient temperature includes daily average. Temperature change rate Power consumption includes: peak hour percentage, which is the ratio of peak hour power consumption to total power consumption; For example, the speed of motor M1: Data for a certain hour is 1450, 1452, 1448, 1455, 1453, 1449. Calculate the average speed: Further calculation of the standard deviation of rotational speed: ; The data shows that the working temperature fluctuation range is 5℃, and the number of start-stop cycles is 3 times per day. The ambient temperature is the average of the 24-hour data for that day, specifically: ; The rate of temperature change over a certain period of time . Power consumption: 30kWh in a certain hour, 500kWh in daily cumulative power consumption, 200kWh in peak hours (18:00-22:00), and 40% of the total power consumption during peak hours (200 / 500).

[0039] The extraction of time-series features is specifically as follows: For motor M1, with a sliding window of 1 hour, the speed is 1450 r / min (t1) and 1460 r / min (t2) at a certain time interval. The first-order difference of the speed is then calculated as follows: ; The slope of the trend is calculated as follows: ; Associating energy consumption with time-series characteristics of operation: Given a power consumption of 30 kWh, an average rotational speed of 1450 r / min, and a running time of 1 hour, calculate the power consumption per unit rotational speed as follows: ; Extracting cross-correlation features specifically involves associating the device with its environment. For example, given that the operating temperature of motor M1 is 45℃ and the ambient temperature is 25℃, the difference between the ambient temperature and the operating temperature is obtained. Equipment density is 0.04 units / m², power consumption per unit area = 500 / 500 = 1 kWh / m², equipment density - power consumption per unit area ratio = 0.04 / 1 = 0.04; the slope of the speed-power consumption fitting is obtained through linear regression: The duration of high-speed (speed > 1400 r / min) periods with ambient temperatures exceeding 28°C accounted for 30%.

[0040] Finally, the feature data is standardized to obtain a time-consistent, error-free structured dataset. The aforementioned feature data is then normalized to form a multi-dimensional matrix. The fused features are then normalized as follows: ; For example: if the average rotational speed ranges from 1400 to 1500 r / min, then ; If the ambient temperature range is 20-30℃, then ; Multi-source data fusion model construction: By fusing feature data, an intelligent fusion model is built. Further intelligent fusion models include: feature-level fusion and decision-level fusion. First, the multi-source feature data is aggregated, and core features for fusion are selected from the feature data of equipment, environment, and energy consumption sub-models to form a unified feature set, including: Equipment characteristics (7 dimensions): mean speed, standard deviation of speed, temperature fluctuation range, number of start-stop cycles, first-order difference of speed, trend slope, temperature difference; Environmental characteristics (5 dimensions): daily average ambient temperature, temperature change rate, duration of high temperature, equipment density, and percentage of high temperature during high-speed periods; Energy consumption characteristics (3 dimensions): hourly power consumption, peak hour percentage, and power consumption per unit speed.

[0041] Align the three types of features by timestamp (e.g., one fused feature vector per hour) to form a fused input matrix. , where n is the number of samples and 15 is the total dimension; Feature-level fusion is based on graph attention network construction, which maps the fused features to graph nodes and edge weights; Furthermore, the nodes are hourly fused feature vectors, such as node... The 15-dimensional features correspond to time t; the edge weights represent the temporal correlation between nodes, calculated using cosine similarity. ,in Let be the feature vector at time t. The higher the similarity, the greater the edge weight. Taking the above 4-hour workshop data as an example, the fused feature vector for each hour is used as a node in the graph. There are 4 nodes here (denoted as...). arrive (These correspond to 10:00-13:00 respectively. Each node contains three types of characteristics: device, environment, and energy consumption, for example...) The equipment features include 7 items such as an average rotational speed of 1451.2 r / min, 5 items such as an ambient temperature of 25.5℃, and 3 items such as a power consumption of 28 kWh.

[0042] The edge weights reflect the degree of connection between nodes, calculated using cosine similarity. and For example, their feature vectors, after normalization, The expression is [0.51, 0.25, 0.4, ..., 0.3]. The values ​​are [0.50, 0.23, 0.42, ..., 0.32]; finally, the weights are obtained. Similarly, calculate It is 0.95. It is 0.94. These non-adjacent nodes have a low weight, 0.88.

[0043] Further, the model parameters are initialized. The input layer receives a 15-dimensional normalized feature vector (7 device features + 5 environmental features + 3 energy consumption features). The hidden layer dimension is set to 64, and the weight matrix is ​​randomly initialized, with each element fluctuating slightly around 0 (e.g., conforming to a normal distribution with a mean of 0 and a standard deviation of 0.01). The attention parameter is a 128-dimensional vector (because two 64-dimensional vectors will be concatenated later), with initial values ​​randomly generated. The number of attention heads is 2, meaning that two sets of attention are computed in parallel, and the results are finally concatenated.

[0044] Further calculate the attention coefficient, here using Taking attention to neighboring nodes as an example, this is processed through a hidden layer. The 15-dimensional feature vector is transformed into a 64-dimensional vector. ;Will vector Each with the neighbors , vector , Concatenate the vectors to obtain two 128-dimensional vectors; then use the LeakyReLU activation function to process the product of the attention parameters and the concatenated vectors to obtain... right The score was approximately 0.85, for The score is approximately 0.78; finally, the score is processed using the softmax function to calculate... right The weights are: Similarly, for Its weight is approximately 0.48.

[0045] Furthermore, update the node features. The new feature vector is obtained by aggregating neighbor features and processed using the ReLU activation function, specifically: This yields an updated 64-dimensional vector. Similarly, all node features are updated in the same way, and finally the results of the two attention heads are concatenated to obtain a 128-dimensional feature vector. Furthermore, the model integrates the fusion features from 14:00 (such as the average rotational speed of 1451.5 r / min and the ambient temperature of 27.5℃, which have been normalized) through an attention mechanism. and Based on the characteristics of the model, the predicted energy consumption is 32.2 kWh, which is 0.2 kWh less than the actual energy consumption of 32 kWh, indicating that the model is effective.

[0046] Decision-level fusion obtains the device sub-model output through LSTM. Obtain the environment sub-model output through GBDT Energy consumption sub-model obtained through ARIMA model ; By acquiring normalized equipment feature data, environmental feature data, and energy consumption feature data, the equipment sub-model output is obtained. Environment submodel output Energy consumption sub-model output ; Further weighted average fusion requires calculating historical average errors. For example, the average prediction error of the equipment sub-model over the past 24 hours is 2kWh, the average error of the environmental sub-model during the same period is 3kWh, and the average error of the energy consumption sub-model during the same period is 4kWh. The weights are inversely proportional to the error. Calculate the corresponding sub-model weights using the formula: ,in This is the model error; Furthermore, we obtained , , The final output is the predicted energy consumption value. .

[0047] Based on the comparison of the two types of fusion results, the feature-level fusion predicts an energy consumption of 32.2 kWh at 14:00, while the decision-level fusion specifically predicts 35 kWh from the equipment sub-model, 32 kWh from the environment sub-model, and 33 kWh from the energy consumption sub-model, resulting in a fusion of 33.6 kWh. The target energy consumption is 30 kWh, therefore the feature-level deviation is 2.2 kWh, and the decision-level deviation is 3.6 kWh. Energy efficiency optimization strategy design: By integrating the output data of the model, an executable energy efficiency optimization strategy is obtained. Both fusion studies show that environmental factors (high temperature) are the primary source of deviation (60% at the feature level and 44% at the decision level), while equipment parameters (speed) are secondary factors (30% at the feature level and 45% at the decision level). This demonstrates a positive correlation between ambient temperature and energy consumption; higher temperatures lead to higher energy consumption. It also shows a correlation between speed and energy consumption; higher speeds result in higher energy consumption.

[0048] The environmental factors are adjusted first. Based on the relationship between feature-level fusion temperature and energy consumption, the target temperature of 24℃ is deduced. Based on the relationship between decision-level fusion temperature and energy consumption, the target temperature of 21.7℃ is deduced. The compromise value of 23℃ is taken (within the normal operating temperature range of the equipment).

[0049] Furthermore, based on the relationship between rotational speed and energy consumption in feature-level fusion, the target rotational speed of 1430 r / min is deduced; similarly, based on decision-level fusion, the target rotational speed of 1420 r / min is deduced; and a compromise value of 1425 r / min is taken (higher than the minimum rotational speed of the equipment of 1400 r / min).

[0050] The further optimized strategy is as follows: turn on the ventilation fan to reduce the ambient temperature from 27.5℃ to 23.5℃; adjust the motor parameters to reduce the speed from 1451.5r / min to 1425r / min, while maintaining a load rate of 75%.

[0051] Model validation and feasibility analysis: The proposed method is compared with traditional methods using real-world scenario data and public datasets to ensure its feasibility. It should be noted that the execution cost C is calculated based on the adjusted strategy. ,in , and These represent the costs of implementing each optimization strategy, if If so, it means that the feasibility is met.

[0052] Using a 2kW fan running for 2 hours daily (off-peak electricity price 0.5 yuan), the cost is 2 yuan; energy consumption is reduced by 3kWh (peak electricity price 1.2 yuan), resulting in a profit of 3.6 yuan, and a net profit of 1.6 yuan / day; the speed is reduced from 1451.5r / min to 1425r / min, reducing energy consumption by 2.06kWh per day, resulting in a profit of 2.47 yuan / day. The solution is feasible and achieves the goal of saving energy and costs.

[0053] Data management and iteration involves real-time storage and management of various types of data throughout the process, summarizing the core innovations of the method, and planning the iterative direction of the method based on feedback from actual applications and technological developments to continuously improve energy efficiency optimization.

[0054] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, 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 within the protection scope of the present invention.

Claims

1. An energy efficiency optimization method based on multi-source data fusion, characterized in that, include: S1. Multi-source data acquisition: By designating a specific scenario as the target scenario for energy efficiency optimization, multi-source data in that scenario is acquired, including: equipment operation sensor data, scenario environment data, and energy consumption data. S2. Data preprocessing and feature engineering: By preprocessing the collected multi-source data, standardized data that has been cleaned, heterogeneous unified, and spatiotemporally aligned is obtained. Feature data is obtained by extracting features from the standardized data. S3. Multi-source data fusion model construction: By fusing feature data, an intelligent fusion model is constructed. The intelligent fusion model includes feature-level fusion and decision-level fusion; wherein, feature-level fusion is constructed based on graph attention networks. The specific steps of the decision-level fusion are as follows: Sub-models are established for equipment, environment, and energy consumption data respectively, and the outputs of each sub-model are obtained. Weights are allocated based on the accuracy of the sub-models using the inverse error method. , The final output is the predicted energy consumption value. ; S4. Energy efficiency optimization strategy design: By integrating the output data of the model, an executable energy efficiency optimization strategy is obtained. The specific steps of the energy efficiency optimization strategy are as follows: S1.1: Comparison of predicted energy consumption values With target energy consumption The deviation was obtained. And further, it can be used to judge whether energy consumption is reasonable based on the deviation; S1.2: If unreasonable, trace the source of the deviation, combine the sub-model output to deduce the key influencing factors, and calculate the sub-model's contribution to the deviation percentage. Specifically: ,like If the contribution deviation accounts for more than 50%, the equipment operating parameters should be optimized. If the contribution deviation accounts for more than 30%, environmental regulation should be optimized. If the contribution deviation exceeds 20%, energy dispatch will be optimized. S1.3: Based on the contribution deviation ratio of the sub-models, corresponding optimization strategies are generated, namely: equipment parameter optimization, environmental control optimization, and energy dispatch optimization. Specifically, equipment parameter optimization involves: based on the correlation model... The target rotational speed within the safety threshold is obtained. And reduce the number of start-stop cycles; environmental control optimization is applicable to the dominant environmental deviation, that is, temperature T is positively correlated with y, then the target temperature must meet the following requirements. And satisfy Within the normal operating range; energy dispatch optimization, applicable to energy consumption time sequence deviation, combined with peak and valley electricity pricing, to transfer high energy-consuming processes to valley hours; S5. Model validation and feasibility analysis: Through real-world scenario data and public datasets, the proposed method is compared with traditional methods to ensure its feasibility. S6. Data Management and Iteration: By storing and managing various types of data throughout the process in real time, we summarize the core innovations of the method, plan the iterative direction of the method based on feedback from actual applications and technological development, and continuously improve the energy efficiency optimization effect.

2. The energy efficiency optimization method based on multi-source data fusion according to claim 1, characterized in that: The multi-source data specifically refers to: The equipment operation sensor data includes: motor speed obtained through the motor's built-in speed sensor, and running time, number of start-stop cycles, and operating temperature collected through the equipment's built-in sensors; Scene environment data includes: ambient temperature obtained by uniformly distributing temperature sensors within the scene, and scene device density obtained by considering scene area and number of devices; Energy consumption data includes electricity consumption obtained through smart meters.

3. The energy efficiency optimization method based on multi-source data fusion according to claim 1, characterized in that: The preprocessing of the collected multi-source data specifically includes: data cleaning, including: Motor speed, using The rule identifies anomalies. For missing data, it imputes the missing data by taking the average of the data points adjacent to the missing point. The specific outlier determination is as follows: ,in The mean, Standard deviation; For operating temperature, data exceeding the equipment threshold are considered extreme values ​​and are replaced with the average value of the same model. Missing data is filled with linear interpolation. Specifically: ； Where t is the current missing time point, t-1 is the previous time point, and t+1 is the next time point. For the temperature data at the next time point, This refers to the temperature data from the previous time point; Ambient temperature fluctuations and noise are corrected using a weighted average; sensor failures are filled using a historical Gaussian distribution model. Equipment density is recalculated after deduplication of equipment numbers. The density is the ratio of the number of equipment to the scene area. Power consumption is assessed by detecting abrupt changes using a sliding window and replacing the abrupt change value with the median value within the window. If an entire segment is missing, its theoretical power consumption needs to be calculated, specifically as follows: ； in, Motor power, Runtime; The format standardization includes unifying the timestamp to "yyyy-MM-dd-HH:mm:ss"; converting all data types to floating-point; and integrating the structure into a table of "timestamp + device ID + data type + value". Spatiotemporal alignment includes time alignment, which uses device data frequency as a benchmark and interpolates low-frequency data; power consumption is divided into equal intervals; and spatial alignment, which binds environmental data according to the region where the device is located. Feature extraction includes: basic statistical features, time-series features, and cross-correlation features. Specifically, the basic statistical features extracted include: motor speed, specifically the hourly average. Standard deviation Peak duration is the ratio of peak duration to total duration; ambient temperature includes daily average. Temperature change rate Power consumption includes: peak hour percentage, which is the ratio of peak hour power consumption to total power consumption; The extraction of time-series features specifically involves: first-order difference of rotational speed. Trend slope Power consumption per unit speed ; The specific cross-correlation feature is: environmental-working temperature difference. Speed-energy consumption fitting slope x represents rotational speed, and y represents power consumption; A time-consistent, error-free structured dataset is obtained through data standardization. The aforementioned feature data is then normalized to form a multidimensional matrix. .

4. The energy efficiency optimization method based on multi-source data fusion according to claim 1, characterized in that: The model validation and feasibility analysis method specifically involves calculating the execution cost C based on the adjusted strategy. ,in , and These represent the costs of implementing each optimization strategy, if If so, it means that the feasibility is met.

5. The energy efficiency optimization method based on multi-source data fusion according to claim 1, characterized in that: The data management and iteration specifically involves: constructing an associated index using a unified timestamp and device ID to form a complete chain of "raw data - feature data - model output - optimization strategy", storing it in the database, and making dynamic adjustments in real time.

6. The data management and iteration according to claim 5, characterized in that: The dynamic adjustment specifically refers to: after executing the optimization strategy, real-time monitoring of actual energy consumption. ,like Then, the correlation model is refitted, and the parameters are iteratively optimized.

Images