An optimization method based on industrial energy consumption big data model

Abstract

The present invention provides an optimization method based on an industrial energy consumption big data model, comprising S1, obtaining a benchmark data set; S2, establishing a target function with the goal of minimizing the total energy consumption per unit industrial production time, and completing the construction of an energy consumption prediction model embedding layer; S3, collecting a real-time energy consumption data set O from a monitoring site j set in the current industrial production cycle. j Enrich the data background and influencing factors, and achieve real-time calibration and display of the fine-tuned model by establishing a contribution matrix; S4, forming a set of energy consumption points. The present invention constructs a bag-of-words model by formulating constraint conditions, which can ensure that the construction and output of the bag-of-words model can meet the constraint requirements of the optimization scheduling model. The obtained model input data is text data with constraints, which ensures that the subsequent energy consumption model training and prediction process not only considers the statistical characteristics of the energy consumption data, but also reduces the computing power constraints in the prior art.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial energy consumption identification and analysis, and specifically to an optimization method based on an industrial energy consumption big data model. Background Art

[0002] At present, the technologies for decarbonization and carbon reduction mainly include the following:

[0003] 1. Renewable energy technology: such as solar energy, wind energy, etc., by using natural energy to replace fossil fuels to reduce carbon emissions. However, renewable energy has problems such as intermittent and instability, and needs to be used in conjunction with energy storage technology. 2. Carbon capture and storage technology: by capturing carbon dioxide from the atmosphere or emission sources and storing it underground or on the seabed to prevent it from entering the atmosphere. However, this technology is still in the development stage, the cost is high, and the safety of long-term storage has not been fully verified. 3. Energy-saving and energy efficiency improvement technology: by improving energy efficiency in industries such as industry, construction, and transportation, reduce energy consumption and carbon emissions. However, this requires a lot of capital investment and technical support, and it is difficult to implement.

[0004] Patent document CN109396576B discloses an optimization decision system and decision method for stability and energy consumption state of EDM based on deep learning. It realizes intelligent control of EDM process through a series of modules such as data preprocessing, feature parameter mining, cluster analysis, database construction, statistical feature acquisition, deep learning prediction and optimization decision. Its defect is that the performance of the system proposed by this technology is highly dependent on data quality. If the input data is biased or inaccurate, it will affect the final decision effect. In addition, the training of deep learning models may require a lot of computing resources and time, which is a challenge in a resource-constrained environment. The system has high requirements for real-time performance. If the delay of model prediction and decision is large, it cannot meet the needs of real-time control. In addition, the generalization ability of the system may be limited by the diversity of training data. If the training data cannot cover all possible processing situations, the system may perform poorly when facing unknown situations. Finally, the complexity of the system may lead to difficulties in maintenance and updating, especially when the algorithm needs iterative optimization.

[0005] Patent CN117273440A discloses an engineering construction Internet of Things monitoring management system and method based on deep learning. By acquiring and processing multi-dimensional data sets in the construction process in real time, it realizes comprehensive monitoring and accurate prediction of the construction status. The system adopts advanced data processing technology, including denoising, feature extraction and data fusion, to improve data quality. At the same time, using deep learning models and knowledge distillation methods, the system can evaluate construction risks and classify anomalies, and promptly discover and respond to potential problems. Its defects are: the methods based on deep learning neural networks mainly include two model algorithms: neural networks based on dual streams and neural networks based on three-dimensional convolution. The neural network based on three-dimensional convolution uses a three-dimensional convolution neural network instead of a two-dimensional convolution neural network in the recognition of operation and maintenance behavior. Although this approach can capture more spatiotemporal information, compared with two-dimensional convolution, the increase in computational complexity brought by three-dimensional convolution is exponential, and it is difficult to meet real-time requirements.

[0006] Based on this, it is necessary to design a more universal and rapid industrial energy consumption identification and analysis technology to effectively improve the accuracy of identifying energy consumption points and energy consumption environment related factors, reduce the amount of calculation, and ensure real-time performance. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention aims to provide an optimization method based on an industrial energy consumption big data model. In order to achieve the above object, the present invention is implemented through the following technical solutions: an optimization method based on an industrial energy consumption big data model, comprising:

[0008] S1. Collect historical data and form a benchmark data set based on the energy consumption data accumulated in the past industrial production cycle;

[0009] S2. Construct an energy consumption prediction model. S2-1. Consider improving energy efficiency and optimizing load distribution in industrial production, establish an objective function with the goal of minimizing the total energy consumption within a unit of industrial production time t, and formulate constraints to ensure stable and efficient operation of the system and meet the optimal scheduling of the energy consumption prediction model; S2-2. Obtain a data set of all Shell command blocks related to energy consumption data, and construct a constraint association matrix to ensure that each command block is associated with the corresponding constraint to meet the system operation requirements; Use a tokenizer to segment the Shell command block, build a dictionary based on the segmentation results, and assign a unique index to each word. At the same time, calculate its frequency in the command block to realize text vectorization, and complete the construction of the embedding layer of the energy consumption prediction model to provide a data basis for energy consumption prediction and optimization; S2-3. Based on a deep learning framework, use an embedding layer to convert word indexes into embedding vectors, and construct a neural network model including a convolutional layer, an LSTM recurrent layer, and a fully connected layer to predict energy consumption in industrial production processes;

[0010] S3, fine-tuning and calibrating the energy consumption prediction model, S3-1, through the real-time energy consumption data set O collected from the monitoring site j set in the current industrial production cycle j Enrich the energy consumption data background and energy consumption influencing factors, establish a contribution matrix, analyze and identify the energy consumption equipment characteristics and environmental factors that have the greatest impact on energy efficiency and load, and determine them as the characteristic variables that need to be fine-tuned in the energy consumption prediction model in the future; S3-2, use the BERT Fine-tuning method to fine-tune the selected characteristic variables and save the final model parameters to ensure that the model achieves optimal performance on specific energy consumption prediction tasks;

[0011] S3-3. Build a React fast framework to achieve real-time calibration and display of the fine-tuned model.

[0012] S4. Create an energy consumption database to store the prediction data of the energy consumption prediction model, identify and mark energy consumption points, and thus form an energy consumption point set.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] 1. Compared with the prior art, the present invention constructs a bag-of-words model by formulating constraint conditions. When processing large-scale industrial energy consumption data, this method can ensure that the construction and output of the bag-of-words model can meet the constraint requirements of the optimization scheduling model. The obtained model input data is text data with constraints, which ensures that the subsequent energy consumption model training and prediction process not only considers the statistical characteristics of the energy consumption data, but also reduces the computing power constraints in the prior art. At the same time, the fused feature set is used to train the LSTM model. Such a model can not only capture the dynamic characteristics of the time series, but also understand the operation mode and equipment status reflected by the collected text data;

[0015] 2. By establishing a contribution matrix C, a quantitative measure is provided for the impact of each device in the industrial production process on each monitoring site. The contribution matrix established by the present invention can determine the uncertainty in equipment operation and the volatility of the production environment, such as equipment failure, changes in raw material quality or random fluctuations in operating conditions, and use this as the characteristic variable that needs to be adjusted, and feed back to the previous text data with constraints, so as to achieve targeted adjustments based on different energy consumption data and constraints, and enable the energy consumption prediction model to more accurately learn and predict energy consumption patterns. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The disclosure of the present invention is described with reference to the accompanying drawings. It should be understood that the drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In the accompanying drawings, the same reference numerals are used to refer to the same components. Among them:

[0017] Figure 1 A schematic diagram of an optimization method based on an industrial energy consumption big data model proposed in an embodiment of the present invention, for evaluating and identifying carbon reduction energy consumption points of a factory in the entire industrial production;

[0018] Figure 2 The figure is a schematic diagram of the steps of an optimization method for an industrial energy consumption big data model proposed in one embodiment of the present invention. DETAILED DESCRIPTION

[0019] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific implementation modes and drawings are only exemplary descriptions of the technical solution of the present invention, and should not be regarded as the whole of the present invention or as a limitation or restriction to the technical solution of the present invention.

[0020] The present invention is further described in detail below with reference to the accompanying drawings, but is not intended to limit the present invention.

[0021] As an understanding of the technical concept and implementation principle of the present invention, the main purpose of the present invention is to solve the following problems existing in the energy consumption model analysis of the prior art:

[0022] Energy consumption behavior recognition is completed based on the detection and representation of sampling points through template matching algorithms based on expert experience. However, these features lack certain flexibility and scalability when faced with large-scale data sets.

[0023] Three-dimensional convolution-based neural networks are used instead of two-dimensional convolutional neural networks in operation and maintenance behavior recognition. Although this approach can capture more spatiotemporal information, compared with two-dimensional convolution, the amount of computation brought by three-dimensional convolution increases exponentially, and it is difficult to meet real-time requirements.

[0024] like Figure 1 - Figure 2 As shown, as an embodiment of the present invention, an optimization method based on an industrial energy consumption big data model is proposed. By quantifying the key variable factors that affect energy consumption (system behavior) in the industrial production process, the characteristics of the text data with constraints are formed, and then combined with the characteristics of the time series data in the industrial production cycle to improve the accuracy of energy consumption prediction, and also avoid the problem of insufficient system computing power caused by diverse data. At the same time, the fused feature data set is used to train the LSTM model, so that the model can not only capture the dynamic characteristics of the time series, but also understand the operation mode and equipment status in industrial production reflected by the text data, so that when predicting energy consumption, the equipment's operating instructions (text data) and the equipment's energy consumption readings (time series data) can jointly provide a more comprehensive prediction perspective. Including specific steps:

[0025] S1. Historical data collection

[0026] When step S1 is implemented, a benchmark data set is formed based on the energy consumption data accumulated in the past industrial production cycle. This benchmark data set is used as the basic data for the subsequent energy consumption prediction model. It is preferably obtained through historical industrial production records, archived energy consumption reports, historical project archives, public databases or literature. It should be noted that the collected energy consumption data at least includes environmental data E env 、Electricity data elec , Gas and water consumption data E res , Equipment operation data E ope .

[0027] Among them, environmental data E env Based on the ambient temperature T monitored in industrial production amb (t) and relative humidity H rel (t) Collected: E env ={T amb (t), H rel (t)}; Electric energy data E elec Based on the power consumption of each measurement interval within a specific measurement period n and the length of each measurement interval, the total energy consumption is calculated by accumulating the power consumption in each time interval and multiplying it by the time of the interval to reflect the power usage within a certain period of time: , P elec,i is expressed as the electrical power consumption in the ith measurement interval in watts (W) or kilowatts (kW), It is expressed as the time length of the i-th measurement interval; gas and water consumption data E res Based on the resource flow F of each measurement interval within a specific time period m j and the length of each measurement interval j To collect: ; Equipment operation data E ope Based on the indicator variable ON of each power-on state of the energy-consuming equipment within a specific time period p k , 1 for on, 0 for off, and the running time D of each energy consumption device k Collected: .

[0028] S2. Building an energy consumption prediction model

[0029] S2-1. Considering improving energy efficiency and optimizing load distribution in industrial production, establish the objective function MAX with the goal of minimizing the total energy consumption per unit industrial production time t. ∅, and formulate constraints to ensure stable and efficient operation of the system and meet the optimal scheduling of the energy consumption prediction model.

[0030] Based on the above technical concept, it can be understood that the purpose of defining the objective function is to optimize the energy consumption performance of a unit production day by improving energy efficiency and reducing energy consumption load, ensuring that each category of energy consumption data can be properly considered in the optimization process, forming a comparison between energy efficiency and energy consumption load. Therefore, the objective function needs to consider the total energy consumption E per unit production time t. total , as well as the importance indicators, efficiency indicators and load indicators of various energy consumption data: (1), where N represents the total number of energy consumption data categories; W j It is expressed as the importance index of the j-th type of energy consumption data in the unit industrial production time t. The importance index is determined by the weighted coefficient of the contribution of the type of energy consumption data to the total energy consumption during calculation; E Eff,j It is expressed as the efficiency index of the jth type of energy consumption data, including unit product energy consumption or energy utilization rate or emission reduction efficiency; E Load,j It is expressed as the load index of the j-th type of energy consumption data, including the total energy consumption or the peak energy consumption demand or the environmental impact load; the j-th type of energy consumption data is specifically divided into the following types: environmental data: such as carbon emissions and pollutant emissions; electric energy data: such as electric energy consumption, peak-valley electricity usage ratio, etc., gas and water consumption data: such as gas consumption and water resource consumption; equipment operation data: such as equipment operation time and number of starts and stops.

[0031] At the same time, in order to make the model's predictions and optimization suggestions conform to the operational restrictions and safety standards in actual industrial production, and to ensure that the model is always stable in the face of uncertainty and fluctuations, the constraints formulated must at least include comprehensive energy consumption limit constraints established to ensure system stability and efficient operation, equipment operation efficiency guarantee constraints, nonlinear constraints on the relationship between production efficiency and energy consumption, dynamic management constraints on environmental impacts, and adaptive constraints on energy costs and market price fluctuations.

[0032] It should be noted that in the process of constructing comprehensive energy consumption constraints, it is necessary to consider the consumption of multiple energy and resources, such as electricity, gas, water resources, as well as carbon emissions and pollutant emissions. In order to further ensure the rationality of energy consumption, a series of constraints need to be introduced, including at least: a. The total energy consumption does not exceed the maximum allowable value; b. The sum of electricity consumption and gas consumption does not exceed the maximum allowable value of electricity; c. Water resource consumption does not exceed the maximum allowable value; d. Carbon emission consumption does not exceed the maximum allowable value; e. Pollutant emission consumption does not exceed the maximum allowable value. In the process of constructing equipment operation efficiency guarantee constraints, it is necessary to require that the proportion of the cumulative value of the equipment's operating time within the available time to the total available time must not be lower than a certain set threshold, so as to ensure the effective use of the equipment and the rational allocation of resources. In the process of nonlinear constraints on the relationship between production efficiency and energy consumption, it is necessary to take into account the production efficiency E ffi Possible nonlinear relationship with energy consumption: (2), where m represents the number of energy consumption data categories, such as electric energy data; aj, bj, and cj represent the energy consumption data E of the jth category, respectively. j The nonlinear coefficient is generally used to describe the energy consumption data E j The square term or constant term has an effect on production efficiency E ffi k represents other factors affecting production efficiency E based on monitoring in industrial production ffi The number of factors, such as raw material quality or equipment status; dh and eh are respectively expressed as the number of factors related to the hth factor S h The relevant coefficient is generally used to describe the production efficiency E ffi There is a linear or constant effect; S h It is expressed as the actual quantity of the hth factor, such as the performance index of raw material quality or equipment status; p is expressed as the performance index other than raw material quality or equipment status monitored in industrial production and affecting production efficiency E ffi The number of other nonlinear factors that affect; f l Expressed as the lth nonlinear factor g l Correlation coefficient, nonlinear factor g l Such as environmental factors, operator skills; E θ Expressed as production efficiency E ffiThe minimum threshold is generally the efficiency standard value. In the process of constructing dynamic management constraints on environmental impacts, it is necessary to consider factors that may change environmental impacts over time and energy consumption. In the process of constructing adaptive constraints on energy costs and market price fluctuations, it is necessary to consider factors that may affect energy consumption optimization due to fluctuations in energy prices. Since energy prices, environmental resilience, etc. are basically fixed and generally do not exceed an acceptable range, and the cost of daily maintenance is relatively small, the above-mentioned dynamic management constraints on environmental impacts and price adaptability constraints are ignored in this embodiment.

[0033] In one embodiment of the present invention, in the existing template matching algorithm based on expert experience, energy consumption behavior recognition is mainly based on the detection and representation of sampling points. However, when facing large-scale data sets, these features lack certain flexibility and scalability. Therefore, the present invention adopts a method of designing an action bag-of-words model and integrating the text features generated by the bag-of-words model with the time series features processed by the LSTM model to complete the energy consumption point recognition analysis. The specific implementation process is as follows:

[0034] S2-2. First, the above constraints are converted into textual form through natural language description or set grammatical rules. For example, the comprehensive energy consumption limit can be expressed by "the total energy consumption shall not exceed XXX units"; the equipment operation efficiency guarantee can be expressed by "the efficiency of the energy-consuming equipment 'booster unit' must be maintained above XXX%"; the nonlinear relationship between production efficiency and energy consumption can be expressed by "the energy consumption per unit of product produced during the industrial production cycle shall not exceed XXX units"; the dynamic management of environmental impacts can be expressed by "carbon emissions must be reduced to below XXX standards"; the adaptability of energy costs to market price fluctuations can be expressed by "energy costs must be adjusted within the market fluctuation range of XXX%".

[0035] Secondly, obtain the data set of all Shell command blocks related to energy consumption data, and build a constraint association matrix to ensure that each command block is associated with the corresponding constraint conditions to meet the system operation requirements; use the Tokenizer to segment the Shell command blocks, build a dictionary based on the segmentation results, and assign a unique index to each word. At the same time, calculate its frequency in the command block to realize text vectorization, and complete the construction of the energy consumption prediction model embedding layer to provide a data basis for energy consumption prediction and optimization.

[0036] As an embodiment of the present invention, when a data set D is obtained according to the collected energy consumption data, D = {(c, f(c)) | c∈C}, it is necessary to satisfy that each Shell command block represents a set of instructions for executing a specific task in the system, so as to match the corresponding energy consumption data for each Shell command block, wherein the matching relationship is as follows: (3),

[0037] In the formula, each command block C, c∈C, is associated with a four-element energy consumption data group (denv, delec, dres, dope); d represents the corresponding type of energy consumption data.

[0038] To ensure that each command block is associated with the corresponding constraint when collecting the Shell command block data set, and in the process of building the bag-of-words model, the constraint is always used as part of the optimization target to ensure that the model output not only reflects the text features but also meets the system's operating constraints. Therefore, the constraint association matrix R is constructed by calculating the sum of the influence of each command block on all energy consumption data types S i =∑ m j=1 R ij ×E j , and all command blocks for each constraint C con Satisfaction The determined build, (4), where T k Expressed as constraint C con The threshold value of P k It is represented as the maximum allowed deviation; the constraint association matrix R is an n×m matrix, where n is the number of matrix rows, representing the number of command blocks, and m is the number of matrix columns, representing the number of energy consumption data types. Each column represents a type of energy consumption data E j ; i, j represent indexes from 1 to n, 1 to m respectively; where if the satisfaction of the command block Greater than or equal to the preset satisfaction threshold T k , then the command block is considered to meet the constraint conditions; otherwise, it is considered not to meet the constraint conditions and further adjustment or optimization is required.

[0039] In one embodiment of the present invention, in order to ensure data quality and consistency, it is also necessary to perform data cleaning on the collected data set D after obtaining all data sets of Shell command blocks related to energy consumption data and before building the constraint association matrix, such as removing irrelevant characters or comments or formatting or filling missing values ​​to ensure data consistency and cleanliness, such as: removing irrelevant characters: identifying and deleting special symbols in the data set D that are irrelevant to energy consumption data, such as non-numeric and unit characters, and HTML tags; deleting comments: using string operations to identify and delete the comment part starting with "#" to ensure the purity of the electric energy data; during formatting processing: unifying the reading attributes of environmental data and electric energy data, converting all temperature and power readings to lowercase; at the same time, unifying the use of spaces in water vapor data and equipment operation data, removing redundant spaces and controlling the consistency of spaces; data cleaning: checking missing values ​​of temperature or humidity readings in environmental data, filling them, and deleting abnormal values ​​in electric energy data elec, such as high energy consumption readings, which may indicate erroneous records or equipment failures.

[0040] In step S2-2, in order to effectively use the collected energy consumption data for the machine learning model, it is necessary to vectorize the data set D after data cleaning. The specific process is as follows:

[0041] First, use the tokenizer to segment all Shell command blocks and obtain the word set H that represents the Shell command block. set At the same time, the identifier is extracted from the collected energy consumption data within the unit industrial production time t to obtain the set H representing the energy consumption data env , H elec , H res and H ope , forming the total set H, H =H set ∪H env ∪H elec ∪H res ∪H ope ; Secondly, construct an empty dictionary Q to store the mapping from words to indices, and traverse each word q in the total set H, assigning a unique index i to each word that does not appear in the dictionary Q: Q(q)=i, if , where the index i is initially 1 and increments after each allocation; finally, a vector F with a length equal to the size of the dictionary Q is created for each Shell command block C c , and initialize it to 0, and traverse each word q in the Shell command block C. At this time, if the word q is in the dictionary Q, then in the vector F C The frequency f(q,C) of word q is recorded at the corresponding index i in the , and F is obtained. C [Q(q)]=f(q,C).

[0042] It should be noted that since the generalization ability of existing prediction models is limited by the diversity of training data, newly emerging construction scenarios or special situations may not be effectively handled. For real-time monitoring systems, any delay in prediction and response may lead to untimely risk management. Taking construction projects as an example, if the construction environment undergoes unexpected changes such as extreme weather conditions, the sensor data that the monitoring system relies on may have errors or insufficient update frequency, and the data obtained is limited by data quality and completeness, resulting in inaccurate energy consumption forecasts. For example, if the environmental conditions at the construction site change, such as rising temperature or changing wind speed, the existing system has the problem of being unable to accurately predict energy consumption due to insufficient generalization ability.

[0043] Therefore, the present invention proposes to combine the text features generated by the bag-of-words model with the time series features processed by the LSTM model. It can be understood that the fusion can be achieved by simple data splicing, and then the fused feature set is used to train the LSTM model, so that the trained prediction model can not only capture the dynamic characteristics of the time series, but also understand the operation mode and equipment status reflected by the text data, thereby ensuring the continuity and effectiveness of the entire energy consumption optimization process.

[0044] Based on this, continue to execute S2-3, based on the deep learning framework, use the embedding layer to convert the word index into an embedding vector, and build a neural network model including a convolutional layer, an LSTM recurrent layer, and a fully connected layer to predict the energy consumption in the industrial production process.

[0045] In step S2-3, the specific process of building a neural network model is as follows:

[0046] First, we initially construct a v×g-dimensional embedding matrix G to map each unique word q in the dictionary Q into a g-dimensional vector space to provide rich input features, where v is represented by the number of unique words q in the dictionary Q, and g is represented by the dimension of the embedding vector; secondly, a unique row index i is assigned to each unique word q in the dictionary Q to define the embedding vector of each word q in the embedding matrix G; at the same time, for each word q in the dictionary Q, a g-dimensional vector is obtained by random initialization to capture the semantic features of the word q; again, the g-dimensional vector of each word q is filled into the corresponding row of the embedding matrix G, and G=[g1,g2,g3,...,gv] is obtained to obtain the embedding vector; from this, a convolutional layer is applied to the output of the embedding layer, and a filter is used to identify local patterns and features to construct the convolutional layer; the output of the convolutional layer is used as the input of the LSTM recurrent layer to construct the LSTM unit, and the forget gate, input gate, unit state and output gate are used to update and transfer information; finally, the output of the LSTM recurrent layer enters one or more fully connected layers, and further extracts features through linear transformation and nonlinear activation functions at the same time, and finally generates the prediction result of the model through the preset activation function at the output layer.

[0047] It can be understood that in the above steps, the output of the bag-of-words model is used as the input of the LSTM model. Therefore, the model needs to better capture the patterns and trends in a specific data set to ensure the continuity and effectiveness of the entire energy consumption optimization process. Therefore, the present invention proposes to use the BERT Fine-tuning method to fine-tune the energy consumption data, that is, to fine-tune the model so that it can better adapt to specific tasks or data sets by adjusting the parameters of the model. The specific process is as follows:

[0048] S3, energy consumption prediction model fine-tuning and calibration, including the following steps:

[0049] S3-1, through the real-time energy consumption data set O collected from the monitoring site j set in the current industrial production cycle j ,First, detailed data are collected from industrial production monitoring sites, including ,energy consumption efficiency, resource utilization efficiency and equipment operation status. ,These data are simulated by constructing a dialogue data set in the ,query chain format to simulate the actual user query scenario, thereby ,enriching the data background and influencing factors. ,By building a contribution matrix, the impact of each energy consuming equipment on the ,monitoring site is quantified. The energy consuming equipment characteristics and environmental factors that ,have the greatest impact on energy consumption efficiency and load are ,identified and determined as the characteristic variables that ,need to be fine-tuned in the subsequent energy consumption prediction model.

[0050] The specific implementation steps are as follows: First, a dialogue dataset in the query chain format is constructed to simulate the actual user query scenario and complete the energy consumption dataset O j Collect, O j ={(Eenv，j (t),E elec，j (t),E res，j (t),E ope，j (t)) | t=1,2,3,...,T}, where T represents the total number of time points of observation monitoring station j; E env，j (t), E elec，j (t), E res，j (t), E ope，j (t) represent the environmental energy consumption data, electric energy consumption data, gas and water consumption data, and equipment operation data of monitoring site j within unit production time t; secondly, based on historical data, determine the background value b that represents the average performance of all monitoring sites j under standard operating conditions and the intensity factor S that measures the environmental impact and resource consumption of energy-consuming equipment k in the industrial production cycle. k , construct a linear combination of the monitoring results at monitoring site j to reflect the impact of energy-consuming equipment on specific indicators of monitoring site j (such as energy consumption, resource usage, etc.), so as to identify the energy-consuming equipment with the greatest impact on energy consumption and resource efficiency, (5), where C kj is represented as the weight factor of energy-consuming device k to monitoring site j; M is represented as the number of devices at monitoring site j; finally, according to the constructed contribution matrix C, the uncertainty in the operation of energy-consuming device k and the volatility of the production environment are identified and quantified, and the characteristics of energy-consuming device k and environmental factors l that have the greatest impact on energy consumption are selected as the characteristic variables that need to be fine-tuned.

[0051] C=[C kj ]M×N(6),

[0052] (7),

[0053] In the formula, It is expressed as a normalization factor, which is used to adjust the weight factor of energy consuming device k to monitoring site j to ensure that the sum of the weights of all devices is 1. k , H k They are respectively the electricity and water consumption generated by energy-consuming equipment k in the industrial production cycle, M is the number of energy-consuming equipment in the contribution matrix C, ɛ is a small constant to prevent the denominator from being zero; w k It is expressed as the weight of the working efficiency or operating status of the energy consuming equipment k; It is expressed as the adaptability of energy-consuming device k under the current environmental conditions, where T represents the actual ambient temperature, Expressed as the average value of ambient temperature, H rel Expressed as the actual relative humidity, Expressed as the average value of relative humidity, σ T , σHrel They are expressed as the standard deviation of ambient temperature and the standard deviation of relative humidity, representing the range of variation of environmental factors; It is expressed as the response of energy-consuming equipment k to various environmental factors l, where gl(R kl ) is expressed as the influence function of device k on environmental factor l, R kl It is represented by the specific parameters of energy-consuming equipment k on environmental factor l.

[0054] S3-2. Fine-tune the selected feature variables using the BERT Fine-tuning method and convert these variables into an input format acceptable to the BERT model, such as word embedding vectors. During the fine-tuning process, load the pre-trained BERT model and freeze most of the parameters. Only fine-tune its top connection layer. Update the model parameters using the loss function and gradient descent method. Use the benchmark dataset to evaluate the performance of the fine-tuned model and save the final model parameters to ensure that the model achieves optimal performance on specific energy consumption prediction tasks.

[0055] The specific implementation steps are as follows: First, collect the energy consumption data set O j After converting to an ID sequence acceptable to the BERT model, load the pre-trained BERT model and initialize the fine-tuning parameters; secondly, use the loss function Freeze most of the parameters of the BERT model and only fine-tune its top connection layer. In the formula, p(y|x) represents the predicted probability of the BERT model for the label output y given the input data x. The fine-tuning formula is as follows: , where It represents the parameters of the connection layer that are updated after fine-tuning the BERT model. Denoted as the learning rate, It is represented as the gradient of the loss function L with respect to the parameters of the BERT model after fine-tuning, and ℮ is represented as the set of parameters before BERT model fine-tuning; finally, using the formula , train the fine-tuned BERT model, where Represented as the new iterative parameters, It is expressed as the gradient of the loss function L with respect to the current BERT model parameters.

[0056] S3-3. Build a React fast framework to achieve real-time calibration and display of the fine-tuned model.

[0057] The specific implementation steps are as follows: First, use a benchmark dataset to compare the results of the current BERT model by comparing the actual energy consumption data E i,act Energy consumption data E predicted by the model i,pre The error score evaluates the accuracy of the BERT model;

[0058] Secondly, the cross entropy method is used to measure the degree to which the BERT model has mastered the correct data: (8), where y i Represented as the actual energy consumption data E of the i-th energy consumption data point i,act The category or state is 0 or 1. is represented as the predicted probability, λ is represented as the smoothing factor, which is a constant, and N is represented as the total number of energy consumption data points considered during calibration. Again, a set of React components for energy consumption data calibration is created, and data calibration is performed through the fully connected layer logic to calculate each actual energy consumption data E i,act Energy consumption data E predicted by the model i,pre The difference is calculated and the calibration threshold is set according to the difference to perform preliminary calibration. If the difference is less than the set calibration threshold, the calibration intensity is reduced, otherwise, the calibration intensity is increased. Finally, JSX syntax and React's virtual DOM are used for rendering and updating to ensure that the calibration and display of energy consumption data are accurate.

[0059] Based on the above technical concept, it should be noted that in the process of real-time calibration of the fine-tuned model, the created React component receives external energy consumption data through props and stores the energy consumption data in the state of the React component. In this process, only the state update of the React component used for energy consumption data calibration is allowed. The state or props of other components always remain in a read-only state during the training phase and cannot be modified.

[0060] S4. Create an energy consumption database to store the predicted data of the energy consumption prediction model, use the TF-IDF algorithm to set feature keywords, build a tree index to identify and mark energy consumption points, and thus form an energy consumption point set to ensure the accessibility and scalability of the data.

[0061] The technical scope of the present invention is not limited to the contents in the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of ​​the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. An optimization method based on industrial energy consumption big data model, characterized by: Includes steps: S1. Collect historical data Form a benchmark data set based on the energy consumption data accumulated during past industrial production cycles; S2. Building an energy consumption prediction model S2-1. Considering improving energy efficiency and optimizing load distribution in industrial production, establish an objective function with the goal of minimizing the total energy consumption per unit industrial production time t, and formulate constraints to ensure stable and efficient operation of the system and meet the optimal scheduling of the energy consumption prediction model; S2-2, obtain the data set of all Shell command blocks related to energy consumption data, and build a constraint association matrix to ensure that each command block is associated with the corresponding constraint condition to meet the system operation requirements; use the tokenizer to segment the Shell command block, build a dictionary based on the segmentation results, and assign a unique index to each word. At the same time, calculate its frequency in the command block to realize text vectorization, and complete the construction of the energy consumption prediction model embedding layer to provide a data basis for energy consumption prediction and optimization; S2-3. Based on the deep learning framework, the word index is converted into an embedding vector using an embedding layer, and a neural network model consisting of a convolutional layer, an LSTM recurrent layer, and a fully connected layer is constructed to predict energy consumption in industrial production processes. S3. Fine-tuning and calibrating the energy consumption prediction model S3-1, through the real-time energy consumption data set O collected from the monitoring site j set in the current industrial production cycle j Enrich the energy consumption data background and energy consumption influencing factors, establish a contribution matrix, analyze and identify the energy consumption equipment characteristics and environmental factors that have the greatest impact on energy efficiency and load, and determine them as characteristic variables that need to be fine-tuned in the subsequent energy consumption prediction model; S3-2. Fine-tune the selected feature variables using the BERT Fine-tuning method and save the final model parameters to ensure that the model achieves optimal performance on the specific energy consumption prediction task; S3-3. Build a fast React framework to achieve real-time calibration and display of the fine-tuned model; S4, creating an energy consumption database to store the prediction data of the energy consumption prediction model, identifying and marking energy consumption points, thereby forming an energy consumption point set; In step S3-1, the specific process of determining the characteristic variables of the fine-tuning energy consumption prediction model is as follows: First, we construct a dialogue dataset in the form of a query chain to simulate the actual user query scenario and complete the energy consumption dataset O j Collect, O j ={(E env，j (t),E elec，j (t), E res，j (t),E ope，j (t)) | t=1,2,3,...,T}, where T represents the total number of time points at which the monitoring station j is observed; E env，j (t), E elec，j (t), E res，j (t), E ope，j (t) respectively represent the environmental energy consumption data, electric energy consumption data, gas and water consumption data and equipment operation data of monitoring site j within unit production time t; Secondly, based on the historical data, determine the background value b that characterizes the average performance of all monitoring sites j under standard operating conditions and the intensity factor S that measures the environmental impact and resource consumption of energy-consuming equipment k in the industrial production cycle. k , construct a linear combination of the monitoring results at the monitoring site j, , to reflect the impact of energy-consuming equipment on the specific indicators of monitoring site j, so as to identify the energy-consuming equipment with the greatest impact on energy consumption and resource efficiency. In the formula, C kj It is represented as the weight factor of energy consumption device k to monitoring site j; M is represented as the number of devices at monitoring site j; Finally, according to the constructed contribution matrix C, the uncertainty in the operation of energy-consuming equipment k and the volatility of the production environment are identified and quantified, and the characteristics of energy-consuming equipment k and environmental factors l that have the greatest impact on energy consumption are selected as the characteristic variables that need to be fine-tuned, C=[C kj ]M×N; , In the formula, It is expressed as a normalization factor, which is used to adjust the weight factor of energy consuming device k to monitoring site j to ensure that the sum of the weights of all devices is 1. k , H k They are respectively represented by the electricity and water consumption generated by energy-consuming equipment k in the industrial production cycle, M is represented by the number of energy-consuming equipment in the contribution matrix C, ɛ is represented by a small constant to prevent the denominator from being zero; w k It is expressed as the weight of the working efficiency or operating status of the energy consuming equipment k; It is expressed as the adaptability of energy-consuming device k under the current environmental conditions, where T represents the actual ambient temperature, Expressed as the average value of ambient temperature, H rel Expressed as the actual relative humidity, Expressed as the average value of relative humidity, σ T , σ Hrel They are expressed as the standard deviation of ambient temperature and the standard deviation of relative humidity, representing the range of variation of environmental factors; It is expressed as the response of energy-consuming equipment k to various environmental factors l, where gl(R kl ) is expressed as the influence function of device k on environmental factor l, R kl It is represented by the specific parameters of energy-consuming equipment k on environmental factor l.

2. The optimization method based on the industrial energy consumption big data model according to claim 1 is characterized in that: In step S3-2, the specific process of fine-tuning the selected feature variables using the BERT Fine-tuning method is as follows: First, the collected energy consumption data set O j After converting to an ID sequence acceptable to the BERT model, load the pre-trained BERT model and initialize the fine-tuning parameters; Second, use the loss function Freeze most of the parameters of the BERT model and only fine-tune its top connection layer, where p ( y | x ) represents the predicted probability of the BERT model for the label output y given the input data x, and the fine-tuning formula is as follows: , where It represents the parameters of the connection layer that are updated after fine-tuning the BERT model. Denoted as the learning rate, It is represented as the gradient of the loss function L with respect to the parameters of the BERT model after fine-tuning, and e is represented as the set of parameters before fine-tuning the BERT model; Finally, using the formula , train the fine-tuned BERT model, where Represented as the new iterative parameters, It is expressed as the gradient of the loss function L with respect to the current BERT model parameters.

3. The optimization method based on the industrial energy consumption big data model according to claim 1 is characterized in that: In step S3-3, the specific process of building a React fast framework to achieve real-time calibration and display of the fine-tuned model is as follows: First, we use a benchmark dataset to compare the results of the current BERT model by comparing the actual energy consumption data E i,act Energy consumption data E predicted by the model i,pre The error score evaluates the accuracy of the BERT model; Secondly, the cross entropy method is used to measure the degree to which the BERT model has mastered the correct data: , where y i Represented as the actual energy consumption data E of the i-th energy consumption data point i,act The category or state is 0 or 1. It is represented as the prediction probability, λ is the smoothing factor, which is taken as a constant, and N is the total number of energy consumption data points considered during calibration; Next, create a set of React components for energy consumption data calibration, perform data calibration through the fully connected layer logic, and calculate each actual energy consumption data E i,act Energy consumption data E predicted by the model i,pre and adjusting the predicted value according to the difference to set a calibration threshold for preliminary calibration, wherein if the difference is less than the set calibration threshold, the calibration intensity is reduced, otherwise, the calibration intensity is increased; Finally, JSX syntax and React's virtual DOM are used for rendering and updating to ensure accurate calibration and display of energy consumption data.

4. The optimization method based on the industrial energy consumption big data model according to claim 1 is characterized in that: In step S2-1, the objective function is determined by considering the total energy consumption within the unit production time t, as well as the importance index, efficiency index and load index of various energy consumption data: , where MAX ∅ It is represented as the objective function of maximizing energy consumption per unit industrial production time t; N is the total number of energy consumption data categories; W j It is expressed as the importance index of the j-th type of energy consumption data in unit industrial production time t; E Eff,j It is expressed as the efficiency index of the j-th type of energy consumption data; E Load,j It is expressed as the load index of the j-th type of energy consumption data; The constraints formulated include at least comprehensive energy consumption limit constraints established to ensure stable and efficient operation of the system, equipment operation efficiency guarantee constraints, nonlinear constraints on the relationship between production efficiency and energy consumption, dynamic management constraints on environmental impacts, and adaptive constraints on energy costs and market price fluctuations.

5. The optimization method based on the industrial energy consumption big data model according to claim 1 is characterized in that: In step S2-2, when a data set D is obtained according to the collected energy consumption data, D = {(c, f(c)) | c∈C}, it is necessary to satisfy that each of the Shell command blocks represents a set of instructions for executing a specific task in the system, so as to match the corresponding energy consumption data for each of the Shell command blocks, wherein the matching relationship is as follows: , where each command block C, c∈C, is associated with a four-element energy consumption data set (denv, delec, dres, dope); d represents the corresponding type of energy consumption data; The constraint association matrix R is calculated by calculating the sum of the impacts of each command block on all energy consumption data types S i =∑ m j=1 R ij ×E j , and all command blocks for each constraint C con Satisfaction The determined build, , where T k Expressed as constraint C con The threshold value of P k It is represented as the maximum allowed deviation; the constraint association matrix R is an n×m matrix, where n is the number of matrix rows, representing the number of command blocks, and m is the number of matrix columns, representing the number of energy consumption data types. Each column represents a type of energy consumption data E j ; i, j represent indexes from 1 to n, 1 to m respectively; Among them, if the satisfaction of the command block Greater than or equal to the preset satisfaction threshold T k , then the command block is considered to meet the constraint conditions; otherwise, it is considered not to meet the constraint conditions and further adjustment or optimization is required.

6. The optimization method based on the industrial energy consumption big data model according to claim 1 or 5 is characterized in that: In step S2-2, the specific process of implementing the text vectorization of dataset D is as follows: First, use the tokenizer to segment all Shell command blocks and obtain the word set H that represents the Shell command block. set At the same time, the identifier is extracted from the collected energy consumption data within the unit industrial production time t to obtain the set H representing the energy consumption data env , H elec , H res and H ope , forming the total set H, H =H set ∪H env ∪H elec ∪H res ∪H ope ; Secondly, construct an empty dictionary Q to store the mapping from words to indices, and traverse each word q in the total set H, assigning a unique index i to each word that does not appear in the dictionary Q: Q(q)=i, if , where index i is initially 1 and increments after each allocation; Finally, for each Shell command block C, create a vector F with a length equal to the size of the dictionary Q c , and initialize it to 0, and traverse each word q in the Shell command block C. At this time, if the word q is in the dictionary Q, then in the vector F C The frequency f(q,C) of word q is recorded at the corresponding index i in the , and F is obtained. C [Q(q)]=f(q,C).

7. The optimization method based on the industrial energy consumption big data model according to claim 1 is characterized in that: In step S2-3, the specific process of constructing the neural network model is as follows: First, a v×g-dimensional embedding matrix G is initially constructed to map each unique word q in the dictionary Q into a g-dimensional vector space to provide rich input features, where v represents the number of unique words q in the dictionary Q and g represents the dimension of the embedding vector. Secondly, a unique row index i is assigned to each unique word q in the dictionary Q to define the embedding vector of each word q in the embedding matrix G; at the same time, for each word q in the dictionary Q, a g-dimensional vector is obtained by random initialization to capture the semantic features of the word q; Again, fill the g-dimensional vector of each word q into the corresponding row of the embedding matrix G, and obtain G=[g1,g2,g3,...,gv] to obtain the embedding vector; Next, a convolutional layer is applied on the output of the embedding layer and filters are used to identify local patterns and features to build a convolutional layer; the output of the convolutional layer is used as the input of the LSTM recurrent layer to build an LSTM unit and update and pass information using the forget gate, input gate, unit state and output gate; Finally, the LSTM recurrent layer output enters one or more fully connected layers, and further extracts features through linear transformation and nonlinear activation function at the same time, and finally generates the prediction result of the model through the preset activation function in the output layer.

8. The optimization method based on the industrial energy consumption big data model according to claim 3 is characterized in that: During the real-time calibration of the fine-tuned model, the created React component receives the external energy consumption data through props and stores the energy consumption data in the state of the React component. In this process, only the state update of the React component used for energy consumption data calibration is allowed. The state or props of other components can only be read during the training phase and cannot be modified.

Images