Load management system and method based on plastic product industry load model

Abstract

The invention discloses a load management system based on a load model in the plastic product industry. The load management system comprises a load time sequence model building module which classifies and marks equipment according to operation behaviors of the equipment to obtain a load time sequence model; the data acquisition and fusion module acquires power, current, voltage and start-stop state data of equipment in real time through a monitoring point position, and generates a real-time load data set in combination with real-time production state data; the load prediction and adjustability evaluation module generates a load prediction curve by using a long short-term memory network, generates an adjustability score by adopting a support vector machine model, and determines an adjustable process list in combination with real-time production state data; and the scheduling execution module determines candidate adjustment equipment according to the load prediction curve, the adjustability score and the adjustable process list, constructs a multi-objective optimization function through a multi-objective optimization algorithm, and generates a load adjustment strategy by using a particle swarm optimization algorithm to execute scheduling on the equipment. According to the invention, the refinement level of load management is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of material product production load management, and in particular to a load management system and method based on a load model of the plastic products industry. Background Art

[0002] The plastics products industry, a vital component of the manufacturing sector, is widely used in packaging, automotive, electronics, and electrical appliances. Its production processes are complex, and it incorporates a wide variety of energy-consuming equipment. Its load characteristics exhibit significant periodicity and volatility. As production scale expands and energy costs continue to rise, companies in the plastics products industry are generally faced with high energy consumption and difficulty in implementing refined load management during production.

[0003] Traditional load management methods in the plastics industry rely heavily on manual judgment or extensive management based on fixed load control strategies. These methods lack precise analysis of equipment operating characteristics and real-time production status feedback, making it difficult to accurately predict future equipment loads and flexibly formulate effective equipment control strategies. This management model not only wastes energy but can also lead to delayed order delivery and the risk of equipment malfunctioning due to inaccurate load forecasts and inappropriate control strategies.

[0004] Existing load forecasting methods primarily focus on general power load forecasting, failing to fully consider the complex process flows and phased nature of the plastics industry's production processes. They also overlook the diverse operating characteristics of various equipment within the industry, resulting in limited forecast accuracy. Furthermore, existing load control strategies typically focus on single-objective optimization, overlooking key factors such as electricity costs, capacity losses, and response default risks. This makes it difficult to meet the power load requirements and production capacity stability needs of plastics companies. Summary of the Invention

[0005] The purpose of the present invention is to provide a load management system based on the load model of the plastic products industry. The present invention improves the refinement level of load management and reduces energy consumption and response risks.

[0006] To achieve this purpose, the present invention designs a load management system based on the plastics industry load model, which includes:

[0007] The load time series model construction module is used to obtain the equipment operating behavior based on the power curve, operating period and start-stop frequency of the equipment in each process of the factory process. The equipment is classified according to the equipment operating behavior, and the classification results and corresponding equipment operating behavior are annotated to obtain the load time series model.

[0008] The data acquisition and fusion module is used to determine the monitoring points of the equipment in each process of the process according to the load time series model, collect the power, current, voltage and start-stop status data of the factory equipment in real time through the monitoring points to obtain real-time equipment operation data, and generate a real-time load data set in combination with the real-time production status data;

[0009] The load forecasting and adjustability assessment module is used to generate a load forecast curve by training the historical operating load series of each plant equipment through a long short-term memory network based on the real-time load data set; use a support vector machine model to score the equipment adjustability based on the historical operating data of the plant equipment and generate an equipment adjustability score; and determine the equipment adjustable process list based on the equipment adjustability score and combined with real-time production status data;

[0010] The scheduling execution module is used to determine candidate regulating equipment based on the load forecast curve, equipment adjustability score and equipment adjustable process list, and construct a multi-objective optimization function that includes electricity expenditure, production capacity loss and response default risk through a multi-objective optimization algorithm. Based on the candidate regulating equipment and the multi-objective optimization function, a load regulation strategy that includes equipment regulation objects, execution time period and load change is generated through the particle swarm optimization algorithm, and factory equipment is scheduled according to the load regulation strategy.

[0011] Preferably, the types of factory equipment in the plastic products industry include twin-screw extruders, injection molding machines, internal mixers, coating rollers and foaming machines; the process flow includes extrusion molding, injection molding, internal mixing and calendering, foaming molding and wet and dry methods; the equipment is classified according to its operating behavior to obtain the classification result as rigid load or adjustable load.

[0012] Preferably, the specific process of constructing the load time series model is:

[0013] Obtain the equipment list and process parameters corresponding to each process step in the plastic products industry; based on historical equipment operation data, extract the power change curve of each device during continuous operation, and construct the power consumption time series of each device; based on the power consumption time series, identify the typical operating time period, start-stop cycle and load peak interval of each device, determine the equipment operation continuity and operation volatility to obtain the equipment operation characteristics, classify the equipment into rigid load or adjustable load according to the equipment operation continuity and operation volatility, and classify and label them according to equipment type, operation characteristics and process coupling relationship to obtain a load time series model.

[0014] Preferably, the monitoring points of the equipment in each process of the process are determined according to the load time series model, and the power, current, voltage and start / stop status data of the factory equipment are collected in real time at the monitoring points to obtain real-time equipment operation data. The specific process of generating a real-time load data set in combination with the real-time production status data is as follows:

[0015] Based on the load time series model, monitoring points of equipment in each process step of the process flow are identified. Power monitoring devices and PLC control interfaces are deployed at the monitoring points of the twin-screw extruder, injection molding machine, internal mixer, coating roller, and foaming machine. During system operation, the power monitoring devices periodically collect power, current, voltage, and start / stop status data of each monitoring point to obtain real-time equipment operation data. The production schedule and process progress data are obtained through the production scheduling system. The start time, product batch, and operation section corresponding to each order are parsed using structured query language. The process plan start / stop time nodes and their deviation from the actual operation status are extracted to form production status tags to obtain real-time production status data.

[0016] The real-time equipment operation data is standardized to eliminate abnormal spikes and short-term packet loss points. The Z-score algorithm is used to detect and mark outliers outside the range of three times the standard deviation. The real-time equipment operation data is smoothed by combining local weighted regression. Based on the segmented normalization strategy of the historical power distribution statistics of each type of equipment, the power, current and voltage are scaled according to the maximum value of the interval, and the start-stop status is encoded using 0-1. The power, current, voltage and start-stop status data of the equipment are associated with the production scheduling position and real-time status of each equipment in the production process. By combining the real-time equipment operation data and real-time production status data, a real-time load data set containing timestamp, equipment ID, process number, operation status, production scheduling progress and energy consumption indicators is constructed.

[0017] Preferably, the specific process of generating a load forecast curve by training the historical operating load sequence of each device in the factory through a long short-term memory network based on the real-time load data set is as follows:

[0018] The calculation formula of the long short-term memory network is as follows:

[0019] h t =o t ⊙tanh(f t ⊙C t-1 +i t ⊙tanh(W c [x t ,h t-1 ,s t ]+b c ))

[0020] Among them, h t is the load forecast result in the load forecast curve output by the long short-term memory network at time t; x t is the standardized load training sequence data at time t; h t-1 is the load forecast result of the load forecast curve output by the long short-term memory network at time t-1; Ct-1 is the unit status at time t-1, which represents the accumulated information of the equipment operating status at the previous time t-1 and is used to predict the load change at the next time; t is the output value of the forget gate, which is used to control the unit state C at the previous moment t-1 The proportion of information that needs to be forgotten in t is the output value of the input gate, which is used to control the degree of update of the unit state by the current input information; t is the output value of the output gate, which is used to control the influence of the cell state on the current predicted output; W c The updated weight matrix for LSTM training is used to map device timing features to the update of unit state; b c The bias vector updated for LSTM training is used to adjust the prediction baseline of the LSTM model; t is the attention weight dynamically generated by the process feature attention layer according to the characteristics of different process stages at time t; ⊙ is the Hadamard product operation;

[0021] The steps for training the historical operating load series of each device in the factory through the long short-term memory network in combination with the calculation formula of the long short-term memory network are as follows:

[0022] Based on the power, current, voltage and start / stop status data of each device in the real-time load data set, combined with the real-time production status data of the device, the device load sequence is marked and divided according to the process stage, and the device load training data with process stage marks is generated;

[0023] Normalizing the equipment load training data according to the load characteristic differences of each process to generate a standardized load training sequence;

[0024] Constructing a long short-term memory network comprising an input layer, a long short-term memory layer, a process feature attention layer, and an output layer, taking the standardized load training sequence as input to the input layer, dynamically adjusting the output weight of the long short-term memory layer according to the characteristics of different process stages through the process feature attention layer, and generating a load forecast result by the output layer;

[0025] The standardized load training sequence is input into the long short-term memory network for forward calculation. After obtaining the predicted output, the error between the load prediction result and the actual load is calculated. Based on the error, the weight parameters of each gate of the long short-term memory layer and the attention weight parameters of the process feature attention layer are updated through the back propagation algorithm until the prediction error converges to the preset threshold. Finally, the load prediction curve is generated by training.

[0026] Preferably, a support vector machine model is used to score the equipment adjustment capability based on the historical operation data of the factory equipment and generate an equipment adjustability score. The specific process of determining the equipment adjustable process list based on the equipment adjustability score and combined with the real-time production status data is as follows:

[0027] Based on the historical operation data of factory equipment, the load response records of each device within the preset time are extracted, and a feature vector containing the number of successful responses, the number of failed responses, the average response time, the advance notification time and the start-stop deviation rate is constructed. The support vector machine model is used for training. The support vector machine model uses the presence of successful response actions in the historical operation data of factory equipment as the classification standard. After the training is completed, the construction of nonlinear decision boundaries is realized through kernel function mapping; in the prediction stage, combined with the equipment start-stop state sequence in the real-time load data set, the current state vector is input into the target equipment, and the adjustability score of the target equipment within the specified response window is calculated. The scoring result is integrated with the key process plan in the real-time production status data, and logical constraints are used to determine whether there is adjustment space. The set of equipment with scores higher than the preset threshold and whose current operating status allows adjustment is screened out to determine the list of adjustable processes for the equipment.

[0028] Preferably, the specific process of constructing a multi-objective optimization function including electricity expenditure, capacity loss and response default risk through a multi-objective optimization algorithm is as follows:

[0029] The formula for constructing a multi-objective optimization function including electricity expenditure, capacity loss and F response default risk through a multi-objective optimization algorithm is as follows:

[0030]

[0031] Wherein, F is the comprehensive target value of the multi-objective optimization function; T is the total number of time periods; P t is the electricity price information at time t; E t is the predicted power consumption at time t; C t is the unit capacity loss cost caused by equipment adjustment at time t; L t is the capacity loss at time t; R t D is the unit default cost of device response failure at time t; t is the probability of responding to default at time t; α is the weight coefficient corresponding to electricity expenditure, β is the weight coefficient corresponding to production capacity loss, and γ is the weight coefficient corresponding to the response default risk;

[0032] The multi-objective optimization function is used to comprehensively evaluate the cost effects of different load regulation strategies in terms of electricity expenditure, capacity loss and response default risk; Indicates the sum of the product of unit electricity price and the corresponding power consumption in each forecast period, which is used to reflect the overall electricity expenditure; capacity loss Represents the product of the capacity loss caused by equipment participation in regulation and the unit capacity loss cost; responds to default risk It means estimating the default probability by the historical response reliability, and then quantifying the response default risk by multiplying it by the default penalty cost; the weight parameters α, β and γ corresponding to the above three items are configured according to demand and weighted aggregated.

[0033] Preferably, based on the candidate regulating devices and the multi-objective optimization function, a load regulation strategy including the device regulation object, execution period and load variation is generated by the particle swarm optimization algorithm. The specific process of scheduling the factory equipment using the load regulation strategy is as follows:

[0034] The formula of the particle swarm optimization algorithm is as follows:

[0035]

[0036] Among them, particle i represents a candidate solution of a device adjustment plan, and represents a set of specific device adjustment operations, including which devices, what method to use, and when to perform them; is the changing trend of the search direction of particle i in the velocity solution space at the k+1th iteration; is the changing trend of the search direction of particle i in the velocity solution space at the kth iteration; is the position of particle i at the kth iteration, which represents the combination of equipment start-stop or load reduction and the corresponding execution period; gbest is the individual historical best position of particle i up to the kth iteration, indicating the best solution position found in the history of particle i; k is the global optimal position of the particle swarm up to the kth iteration, indicating the optimal solution position known in the current entire particle swarm; ω is the inertia weight coefficient; c1 is the individual cognitive factor; c2 is the group cognitive factor; r1 is the first random number in the range of [0,1]; r2 is the second random number in the range of [0,1];

[0037] The particle swarm optimization algorithm is used to perform a global search for the adjustment combinations and adjustment periods of various equipment in the factory. The initial particle swarm is composed of multiple start-stop and load reduction combinations. Each particle represents a feasible adjustment strategy, and the particle position encoding includes equipment identification, operation mode and specific time window; for the initial particle swarm, particle swarm individuals corresponding to various equipment in the factory are generated according to the candidate adjustment equipment, and the executable time window is limited in combination with the process scheduling information. Based on the load forecast curve, the load change prediction value of the adjustment combination corresponding to each particle in the selected period is calculated, and the prediction value is substituted into the multi-objective optimization function for fitness calculation. In each round of iteration, the individual optimal position of the particle is obtained by comparing its current fitness with the historical optimal fitness. Comparison is updated, and the global optimal position is dynamically tracked according to the optimal fitness of the group; the speed vector is adjusted by combining the individual cognitive factor and the group cognitive factor, and the fusion guidance of the individual optimal and the global optimal is achieved through weighted summation to realize the update of the load regulation strategy. During the iterative process, when the overall fitness fluctuation of the particle swarm is lower than the set threshold or reaches the maximum number of iterations, it is considered to have converged, and the current global optimal load regulation strategy is output. In combination with the multi-objective optimization function, a load regulation strategy including the equipment regulation object, execution period and load change is selected, that is, the load regulation strategy with the lowest electricity price, the smallest production capacity loss and controllable response default risk, and the load regulation strategy is sent to the factory equipment for execution and scheduling through the PLC control interface.

[0038] A load management method based on a load model for the plastics industry comprises the following steps:

[0039] The equipment operation behavior is obtained based on the power curve, operation period and start-stop frequency of the equipment in each process of the factory process. The equipment is classified according to the equipment operation behavior, and the classification results and the corresponding equipment operation behavior are labeled to obtain the load time series model;

[0040] Determine the monitoring points of the equipment in each process step of the process according to the load time series model, collect the power, current, voltage and start-stop status data of the factory equipment in real time through the monitoring points to obtain real-time equipment operation data, and generate a real-time load data set in combination with the real-time production status data;

[0041] Based on the real-time load data set, a long short-term memory network is used to train the historical operating load series of each device in the factory to generate a load forecast curve; a support vector machine model is used to score the equipment's adjustability based on the historical operating data of the factory equipment and generate an equipment adjustability score; based on the equipment adjustability score and combined with real-time production status data, a list of adjustable processes for the equipment is determined;

[0042] Candidate regulating equipment is determined based on the load forecast curve, equipment adjustability score and equipment adjustable process list. A multi-objective optimization function that includes electricity expenditure, production capacity loss and response default risk is constructed through a multi-objective optimization algorithm. Based on the candidate regulating equipment and the multi-objective optimization function, a load regulation strategy that includes equipment regulation objects, execution time period and load change is generated through the particle swarm optimization algorithm. Factory equipment is scheduled according to the load regulation strategy.

[0043] A computer program product comprises a computer program, wherein when the computer program is executed by a processor, the steps of the above method are implemented.

[0044] Beneficial effects of the present invention:

[0045] The present invention realizes efficient load management of factories in the plastic products industry, can effectively reduce electricity expenses, production capacity loss and response default risks, improve the economic benefits and operational stability of factories in the plastic products industry, and at the same time optimize the production process, ensure the continuity and flexibility of production, realize accurate prediction and intelligent regulation, improve the refinement level of load management, and reduce energy consumption and response risks. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a structural schematic diagram of the present invention;

[0047] Figure 2 Flowchart of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0049] Example 1

[0050] A load management system based on the load model of the plastic products industry, such as Figure 1 As shown, it includes:

[0051] The load time series model construction module is used to obtain equipment operating behavior based on the power curve, operating time period, and start-stop frequency of equipment in each process step of the factory process (collecting operating power time series data of equipment in the plastic products industry factory in the past three months). Equipment is classified according to equipment operating behavior, and the classification results and corresponding equipment operating behavior are labeled to obtain the load time series model. Through detailed analysis of equipment operating behavior, this design can better understand the energy consumption pattern of equipment, thereby providing a basis for optimized scheduling. Through classification and labeling, equipment can be grouped according to energy consumption pattern, simplifying management processes and improving management efficiency. The load time series model provides rich historical data, enabling the prediction model to more accurately capture load variation patterns.

[0052] The data acquisition and fusion module is used to determine the monitoring points of the equipment in each process of the process according to the load time series model (high-precision power monitoring devices and PLC control interfaces are deployed at the monitoring points). The power, current, voltage and start-stop status data of the factory equipment are collected in real time at the monitoring points to obtain real-time equipment operation data, and the real-time load data set is generated in combination with the real-time production status data. This design collects the power, current, voltage and start-stop status data of the factory equipment in real time at the monitoring points, which can fully grasp the real-time operation status of the factory equipment and provide real-time and accurate data support for load management and optimization. By combining the real-time equipment operation data with the production status data to generate a real-time load data set, it can more comprehensively understand the load situation of the plastic products industry factory, timely adjust the load management strategy, and provide a more accurate data basis for load forecasting and optimization.

[0053] The load forecasting and adjustability evaluation module is used to train the historical operating load sequences of each plant device through a long short-term memory network (LSTM) based on the real-time load data set to generate a load forecast curve (used to identify the peak intervals where each plant device may exceed the load threshold (the load threshold is set according to the specific device and actual scenario)); a support vector machine model (SVM) is used to score the equipment's adjustability based on the plant's historical operating data and generate an equipment adjustability score; a list of adjustable process steps for the equipment is determined based on the equipment adjustability score and combined with real-time production status data. This design can effectively capture the time series characteristics of load data through the long short-term memory network, thereby improving the accuracy of load forecasting; and uses the support vector machine model to score the equipment's adjustability and generate an equipment adjustability score, thereby identifying which devices have higher adjustability and providing a basis for optimized scheduling;

[0054] The scheduling execution module is used to determine candidate regulating devices based on the load forecast curve, equipment adjustability score, and equipment adjustable process list. A multi-objective optimization algorithm is used to construct a multi-objective optimization function that includes electricity expenditure, production capacity loss, and response default risk. Based on the candidate regulating devices and the multi-objective optimization function, a particle swarm optimization algorithm (the full name of the particle swarm optimization algorithm is Improved Particle Swarm Optimization, abbreviated as IPSO) is used to generate a load regulation strategy that includes equipment regulation objects, execution time periods, and load changes. Factory equipment is scheduled using the load regulation strategy (according to the load regulation strategy, start, stop, and load reduction instructions are sent to the corresponding equipment in the plastic products industry factory through the PLC control interface, and the equipment response is monitored in real time and the actual load change results are recorded). This design can balance multiple objectives through the multi-objective optimization algorithm to generate the optimal scheduling strategy. The particle swarm optimization algorithm can quickly find the optimal solution, improve the efficiency of scheduling execution, and enhance the economic benefits and operational stability of the plastic products industry factory.

[0055] In the above technical solution, the types of factory equipment in the plastic products industry include twin-screw extruders, injection molding machines, internal mixers, coating rollers and foaming machines; the process flows include extrusion molding, injection molding, internal mixing and calendering, foaming molding and wet and dry processes (the process flows are all five typical process flows in factories in the plastic products industry); the equipment is classified according to its operating behavior to obtain the classification results as rigid loads (equipment with large operating continuity and small volatility is classified as rigid loads) or adjustable loads (equipment with frequent start-stop or periodic fluctuation characteristics is classified as adjustable loads).

[0056] In the above technical solution, the specific process of constructing the load time series model is as follows:

[0057] Obtain the equipment list and process parameters corresponding to each process step in the plastic products industry; based on historical equipment operation data, extract the power change curve of each device during continuous operation and construct the power consumption time series of each device; based on the power consumption time series, identify the typical operating time period, start-stop cycle and load peak interval of each device, determine the equipment operation continuity and operation volatility to obtain the equipment operation characteristics, classify the equipment as rigid load or adjustable load based on the equipment operation continuity and operation volatility, and classify and label them according to equipment type, operation characteristics and process coupling relationship to obtain a load time series model; by analyzing these characteristics in detail, the above design can more accurately identify the load characteristics of the equipment, provide a basis for optimized scheduling, establish a load time series model to improve the accuracy of load forecasting, and thus better support optimized scheduling;

[0058] In the process of constructing the above-mentioned load time series model, in order to achieve accurate construction of the load model, five typical process flows including extrusion molding, injection molding, mixing and calendering, foam molding and dry coating were first selected from plastic product manufacturers to collect information on high-energy-consuming equipment involved in the above processes; the original information obtained not only includes the equipment list, but also covers process parameters such as equipment rated power, process operating temperature and standard operating time to form a structured basic equipment database; the equipment operation data of the past 6 months in the enterprise energy consumption management platform is called, and the corresponding electricity consumption time series is constructed for each device based on the equipment operation log and power curve. The sequence uses minutes as the time granularity to fully characterize the startup, stable operation and shutdown stages of each type of equipment in the complete production cycle, and combines the sliding window algorithm to extract the load change slope, short-term fluctuation amplitude and mean square error for modeling equipment operation stability and power fluctuation. Based on the extracted features, the equipment operation continuity index (such as the ratio of the average stable period duration to the maximum operation time) and volatility index (such as the standard deviation change rate per unit time) are defined, and the typical operation cycle and high-frequency start-stop mode are identified through second-order difference analysis; the design decision rules divide the equipment with large operation continuity and small volatility into rigid loads, and divide the equipment with frequent start-stop or periodic fluctuation characteristics into adjustable loads; in order to further improve the industry adaptability of the load timing model, combined with the coupling strength of each equipment in the process flow (represented by the dependency matrix between process input and output) and load characteristics, all equipment are multi-dimensionally clustered through the hierarchical clustering algorithm, and the process, type and adjustment attributes of the equipment are marked in the clustering results, and finally a load timing model with structured classification characteristics and adjustment feasibility indicators is formed.

[0059] In the above technical solution, the monitoring points of the equipment in each process step of the process are determined according to the load time series model. The power, current, voltage and start / stop status data of the factory equipment are collected in real time at the monitoring points to obtain real-time equipment operation data. The specific process of generating a real-time load data set by combining the real-time production status data is as follows:

[0060] Based on the load time series model, monitoring points of the equipment in each process of the process are identified, and power monitoring devices and PLC control interfaces are deployed at the monitoring points of the twin-screw extruder, injection molding machine, internal mixer, coating roller and foaming machine (multi-channel power monitoring devices with an accuracy of not less than 0.5% are deployed on the twin-screw extruder, injection molding machine, internal mixer, coating roller and foaming machine in the production line of the plastic products enterprise, and are connected to the PLC control interface via the RS-485 bus). During the operation of the system, the power, current, voltage and start-stop status data of each monitoring point are periodically collected by the power monitoring device to obtain real-time equipment operation data; the production schedule and process progress data are obtained through the production scheduling system, and the start time, product batch and operation section corresponding to each order are parsed using structured query language. The start and stop time nodes of the process plan and the deviation information from the actual operation status are extracted to form production status tags to obtain real-time production status data;

[0061] During the data collection process of the above-mentioned power monitoring device, in order to enhance the time synchronization of the data, a unified sampling period (for example, 10 seconds) is used during the implementation process to cache the data of each collection channel in the edge computing gateway, and the time series fusion of the data of different devices is achieved through timestamp alignment and time calibration mechanism;

[0062] Real-time equipment operation data is normalized to remove abnormal spikes and short periods of packet loss. A Z-score algorithm is used to detect and mark outliers outside the range of three standard deviations. Local weighted regression is then used to smooth the data. Based on a piecewise normalization strategy based on historical power distribution statistics for each type of equipment, power, current, and voltage are scaled according to their maximum values. A 0-1 encoding is used for start-stop status. The power, current, voltage, and start-stop status data of each device are associated with its production location and real-time status in the production process. By combining real-time equipment operation data with real-time production status data, a real-time load dataset is constructed that includes timestamps, equipment IDs, process numbers, operation status, production schedules, and energy consumption indicators. This design, through real-time data collection, captures the latest equipment operation information, enabling more reasonable scheduling decisions. Data fusion generates a more accurate real-time load dataset, supporting subsequent load forecasting and optimization. This real-time load dataset is highly timely and consistent, significantly improving the dynamic responsiveness and accuracy of load management strategies.

[0063] In the above technical solution, based on the real-time load data set, the specific process of generating the load forecast curve by training the historical operating load series of each device in the factory through the long short-term memory network is as follows:

[0064] The calculation formula of the long short-term memory network is as follows:

[0065] ht =o t ⊙tanh(f t ⊙C t-1 +i t ⊙tanh(W c [x t , h t-1 , s t ]+b c ))

[0066] Among them, h t is the load forecast result in the load forecast curve output by the long short-term memory network at time t; x t is the standardized load training sequence data at time t; h t-1 is the load forecast result of the load forecast curve output by the long short-term memory network at time t-1; C t-1 is the unit status at time t-1, which represents the accumulated information of the equipment operating status at the previous time t-1 and is used to predict the load change at the next time; t is the output value of the forget gate, which is used to control the unit state C at the previous moment t-1 The proportion of information that needs to be forgotten in t is the output value of the input gate, which is used to control the degree of update of the unit state by the current input information; t is the output value of the output gate, which is used to control the influence of the cell state on the current predicted output; W c The updated weight matrix for LSTM training is used to map device timing features to the update of unit state; b c The bias vector updated for LSTM training is used to adjust the prediction baseline of the LSTM model; t is the attention weight dynamically generated by the process feature attention layer according to the characteristics of different process stages at time t; ⊙ is the Hadamard product operation;

[0067] In the output calculation phase of the long short-term memory network constructed above, process stage semantic information is introduced based on the process feature attention mechanism to enhance the long short-term memory network model's ability to perceive the stage characteristics of load changes; in the hidden state calculation at the current moment, a composite expression method that integrates the gating mechanism and the process attention factor is adopted: the unit state at the previous moment is attenuated by the forgetting gate, and the candidate state update item is constructed by combining the current input and historical state information activated by the input gate; the process feature attention weight at the current moment is concatenated with the input feature and the historical state to jointly participate in the unit state update process, realizing the regulation of the network memory path by the process feature; the output gate is used to gate the activated state information to obtain the predicted state at the current moment. In the above calculation process, the output and state update operations of each gated unit are performed through the Hadamard product method to ensure the consistency of the vector dimension and the stability of the structure;

[0068] The steps for training the historical operating load series of each device in the factory through the long short-term memory network in combination with the calculation formula of the long short-term memory network are as follows:

[0069] Based on the power, current, voltage and start / stop status data of each device in the real-time load data set, combined with the real-time production status data of the device, the device load sequence is marked and divided according to the process stage, and the device load training data with process stage marks is generated;

[0070] Normalizing the equipment load training data according to the load characteristic differences of each process to generate a standardized load training sequence;

[0071] Constructing a long short-term memory network comprising an input layer, a long short-term memory layer, a process feature attention layer, and an output layer, taking the standardized load training sequence as input to the input layer, dynamically adjusting the output weight of the long short-term memory layer according to the characteristics of different process stages through the process feature attention layer, and generating a load forecast result by the output layer;

[0072] The standardized load training sequence is input into the long short-term memory network for forward calculation. After obtaining the predicted output, the error between the load prediction result and the actual load is calculated. Based on the error, the weight parameters of each gate of the long short-term memory layer and the attention weight parameters of the process feature attention layer are updated through the back propagation algorithm until the prediction error converges to a preset threshold (the preset threshold needs to be dynamically adjusted according to data characteristics, model performance requirements and actual application scenarios), and finally the load forecast curve is trained and generated. The above design trains the historical operation load sequence of the equipment through the long short-term memory network to generate a more accurate load forecast curve. The introduction of the process feature attention layer can dynamically adjust the output weight of the long short-term memory network layer according to the characteristics of different process stages, so that the long short-term memory network model can better focus on the key features of different process stages, further improving the accuracy of load forecasting, and normalizing the equipment load training data to eliminate the dimensional differences between different equipment and processes, so that the long short-term memory network model can process various data more fairly, thereby improving the generalization ability of the long short-term memory network model.

[0073] The specific process of training the historical operating load sequence of each device in the factory through the long short-term memory network is as follows: based on the constructed real-time load data set, the operating power, current, voltage and start-stop status data of each device under actual working conditions are extracted from it, and combined with the real-time process label information obtained by the production execution system, each equipment load sequence is marked and divided into stages according to the process stage, and the equipment load training data containing process context semantics is obtained; in view of the actual situation that the load characteristics of different processes in the plastic products industry are significantly different, the training data is normalized and the minimum-maximum normalization (Min-Max Scaling) strategy normalizes each indicator to the range of [0,1], thereby constructing a standardized load training sequence under a unified scale; in terms of neural network structure design, a composite LSTM prediction model is constructed, which includes an input layer, a long and short-term memory layer, a process feature attention layer, and an output layer; the input layer is used to receive the normalized load training sequence, and the LSTM layer is composed of a stack of multiple memory units, each of which contains a forget gate, an input gate, and an output gate to capture the long-term dependency structure and short-term fluctuation characteristics in the load data; in order to further enhance the model's ability to pay attention to process switching nodes and key process stages, a process feature attention layer is introduced after the LSTM output, and the attention weights generated by training are used to monitor the input of the LSTM at each moment. The weighted aggregation is performed to achieve dynamic weighted processing of process stages with different load characteristics, thereby improving the accuracy of modeling sudden loads and periodic fluctuations; during the training of the long short-term memory network model, the mean square error (MSE) is used as the loss function, and the Adam optimizer is selected to execute the parameter backpropagation update strategy; after each round of iteration, the prediction error of the model on the validation set is calculated, and the learning rate is dynamically adjusted to control the convergence speed; when the downward trend of the validation set error tends to be stable and meets the preset convergence threshold, the training is terminated and the parameters are frozen; finally, the long short-term memory network model outputs the load forecast curve for each device in the next 24 hours, with a prediction granularity of 10 minutes per frame, which has the high timeliness and high resolution characteristics required for flexible adjustment strategy decision-making.

[0074] In the above technical solution, a support vector machine model is used to score the equipment's adjustability based on the factory equipment's historical operating data and generate an equipment adjustability score. The specific process of determining the equipment's adjustable process list based on the equipment adjustability score and combined with real-time production status data is as follows:

[0075] Based on the historical operation data of factory equipment, the load response records of each device within the preset time are extracted, and a feature vector including the number of successful responses, the number of failures, the average response time, the advance notification time and the start-stop deviation rate is constructed. The support vector machine model is used for training. The support vector machine model uses the existence of successful response actions in the historical operation data of factory equipment as the classification standard. After the training is completed, the construction of nonlinear decision boundaries is realized through kernel function mapping; in the prediction stage, the current state vector is input to the target device in combination with the equipment start-stop state sequence in the real-time load data set, and the adjustability score of the target device within the specified response window is calculated. The scoring result is integrated with the key process plan in the real-time production status data, and logical constraints are used to determine whether there is room for adjustment. The above-mentioned support vector machine model realizes the construction of nonlinear decision boundary through kernel function mapping, and can process complex and nonlinear data relationships, so as to more accurately classify the adjustment capability of equipment and improve the accuracy of evaluation. By screening out the equipment set with a score higher than the preset threshold and the current operation status allowing adjustment, the equipment adjustable process list can be determined, which provides clear guidance for optimized scheduling and helps to achieve optimized load management under the premise of meeting production needs. By accurately evaluating the adjustment capability of equipment, equipment with high response success rate and short response time is selected for adjustment, which reduces the risk of response failure and improves overall operation efficiency.

[0076] In the above technical solution, the specific process of constructing a multi-objective optimization function including electricity expenditure, capacity loss and response default risk through a multi-objective optimization algorithm is as follows:

[0077] The formula for constructing a multi-objective optimization function including electricity expenditure, capacity loss and F response default risk through a multi-objective optimization algorithm is as follows:

[0078]

[0079] Wherein, F is the comprehensive target value of the multi-objective optimization function; T is the total number of time periods; P t is the electricity price information at time t; E t is the predicted power consumption at time t; C t is the unit capacity loss cost caused by equipment adjustment at time t; L t is the capacity loss at time t; R t D is the unit default cost of device response failure at time t; t is the probability of responding to default at time t; α is the weight coefficient corresponding to electricity expenditure, β is the weight coefficient corresponding to production capacity loss, and γ is the weight coefficient corresponding to the response default risk;

[0080] The multi-objective optimization function is used to comprehensively evaluate the cost effects of different load regulation strategies in terms of electricity expenditure, capacity loss and response default risk; Indicates the sum of the product of unit electricity price and the corresponding power consumption in each forecast period, which is used to reflect the overall electricity expenditure; capacity loss Represents the product of the capacity loss caused by equipment participation in regulation and the unit capacity loss cost; responds to default risk It means estimating the default probability by the historical response reliability, and then quantifying the response default risk by multiplying it by the default penalty cost; the weight parameters α, β and γ corresponding to the above three items are configured according to demand and weighted aggregated.

[0081] In the above technical solution, based on the candidate regulating devices and the multi-objective optimization function, a load regulation strategy including the device regulation object, execution period, and load change is generated by the particle swarm optimization algorithm. The specific process of scheduling the factory equipment using the load regulation strategy is as follows:

[0082] The formula of the particle swarm optimization algorithm is as follows:

[0083]

[0084] Among them, particle i represents a candidate solution of a device adjustment plan, and represents a set of specific device adjustment operations, including which devices, what method to use, and when to perform them; is the changing trend of the search direction of particle i in the velocity solution space at the k+1th iteration; is the changing trend of the search direction of particle i in the velocity solution space at the kth iteration; is the position of particle i at the kth iteration, which represents the combination of equipment start-stop or load reduction and the corresponding execution period; gbest is the individual historical best position of particle i up to the kth iteration, indicating the best solution position found in the history of particle i; k is the global optimal position of the particle swarm up to the kth iteration, indicating the optimal solution position known in the current entire particle swarm; ω is the inertia weight coefficient; c1 is the individual cognitive factor; c2 is the group cognitive factor; r1 is the first random number in the range of [0,1]; r2 is the second random number in the range of [0,1];

[0085] In the formula of the above particle swarm optimization algorithm, the particle swarm optimization algorithm realizes the optimal strategy solution for the equipment adjustment combination and execution period by simulating the information sharing and collaborative search behavior among individuals in the group; in each round of iteration, the algorithm uses the "speed" and "position" of the particle as the control variables, where "speed" represents the trend of the change of the search direction of the particle in the current solution space, and "position" represents a specific equipment start-stop or load reduction adjustment scheme corresponding to the current particle; the next search direction (i.e., speed) of the particle is affected by three factors: first, the speed inertia of the particle itself in the previous iteration, which reflects the search The first is the continuity of the search; the second is the optimal solution position found by the particle in history, called individual optimality, which is used to strengthen the individual's experience learning; the third is the optimal solution position known in the entire particle group, called global optimality, which is used to guide the individual to move closer to the collective cognition; the above three factors act together in the speed adjustment formula, and the influence strength of inertia, individual cognition and group cognition is controlled by weighting factors respectively; in order to improve the global exploration ability and local convergence ability of the algorithm, the process also introduces random variables that obey a uniform distribution to perturb the learning factor in the speed update formula, enhance the diversity of the search path, and avoid falling into the local optimality. As the iteration proceeds, the particles continuously adjust their positions under the guidance of individual experience and group optimality, that is, continuously generate and evaluate new adjustment plans; in each evaluation, the system substitutes the plan represented by the particle into the aforementioned multi-objective optimization function, calculates its comprehensive cost in terms of electricity expenditure, production capacity loss and response default risk, and judges the pros and cons of the plan through the fitness value. Finally, the particle swarm will converge to the optimal solution under the current conditions after multiple iterations, thereby outputting an equipment load adjustment strategy that meets the multi-objective balance requirements;

[0086] The particle swarm optimization algorithm is used to perform a global search for the adjustment combinations and adjustment periods of various equipment in the factory. The initial particle swarm is composed of multiple start-stop and load reduction combinations. Each particle represents a feasible adjustment strategy, and the particle position encoding includes the equipment identification, operation mode and specific time window. For the initial particle swarm, the particle swarm individuals corresponding to each equipment in the factory are generated according to the candidate adjustment equipment. The executable time window is limited in combination with the process scheduling information. Based on the load forecast curve, the load change prediction value of each particle corresponding to the adjustment combination in the selected time period is calculated. The prediction value is substituted into the multi-objective optimization function for fitness calculation. In each round of iteration, the individual optimal position of the particle is obtained. By comparing and updating its current fitness with the historical optimal fitness, the global optimal position is dynamically tracked according to the group optimal fitness; the speed vector is adjusted by combining the individual cognitive factor and the group cognitive factor, and the fusion guidance of the individual optimal and the global optimal is achieved through weighted summation to realize the update of the load regulation strategy. During the iterative process, when the overall fitness fluctuation of the particle swarm is lower than the set threshold or reaches the maximum number of iterations, it is considered to have converged, and the current global optimal load regulation strategy is output. In addition, the load regulation strategy including the equipment regulation object, execution period and load change is selected in combination with the multi-objective optimization function, that is, the lowest electricity price, the smallest production capacity loss and the response to the default risk A controllable load regulation strategy (the strategy includes equipment number, regulation mode (start / stop / load reduction), execution start and end time periods and expected load change values, and is sent to each equipment control terminal through the PLC control interface to achieve closed-loop control at the instruction level), and the load regulation strategy is sent to the factory equipment for execution scheduling through the PLC control interface; the above design can comprehensively evaluate the cost effects of different load regulation strategies in multiple key aspects by constructing a multi-objective optimization function including electricity expenditure, production capacity loss and response default risk. The system can achieve optimal scheduling with the lowest electricity cost, the smallest production capacity loss and controllable response default risk under the premise of meeting production needs. The weight parameters can be configured to flexibly adjust the weights of different objectives according to the actual needs and priorities of the plastic products industry factory, so that the optimization strategy is more in line with the actual operation needs of the plastic products industry factory. The particle swarm optimization algorithm is used to conduct a global search for the adjustment combination and adjustment period of each equipment in the factory, which can quickly find the optimal load adjustment strategy. In each round of iteration, the individual optimal position and the global optimal position of the particle can be dynamically updated. The speed vector is adjusted by combining the individual cognitive factor and the group cognitive factor. The fusion guidance of the individual optimal and the global optimal is achieved through weighted summation, so that the algorithm can gradually approach the optimal solution and improve the optimization efficiency.

[0087] Example 2

[0088] A load management method based on the load model of the plastic products industry, such as Figure 2As shown in the figure, the equipment is classified according to its operating behavior, and the load time series model is obtained after labeling; the power, current, voltage and start-stop status data of the equipment are collected in real time through monitoring points, and the real-time load data set is generated in combination with the real-time production status data; the load forecast curve is generated by the long short-term memory network, the adjustability score is generated by the support vector machine model, and the adjustable process list is determined in combination with the real-time production status data; the candidate adjustment equipment is determined according to the load forecast curve, adjustability score and adjustable process list, and the multi-objective optimization function is constructed through the multi-objective optimization algorithm. The particle swarm optimization algorithm is used to generate the load adjustment strategy to execute the scheduling of the equipment.

[0089] The specific method of load management includes the following steps:

[0090] The equipment operation behavior is obtained based on the power curve, operation period and start-stop frequency of the equipment in each process of the factory process. The equipment is classified according to the equipment operation behavior, and the classification results and the corresponding equipment operation behavior are labeled to obtain the load time series model;

[0091] Determine the monitoring points of the equipment in each process step of the process according to the load time series model, collect the power, current, voltage and start-stop status data of the factory equipment in real time through the monitoring points to obtain real-time equipment operation data, and generate a real-time load data set in combination with the real-time production status data;

[0092] Based on the real-time load data set, a long short-term memory network is used to train the historical operating load series of each device in the factory to generate a load forecast curve; a support vector machine model is used to score the equipment's adjustability based on the historical operating data of the factory equipment and generate an equipment adjustability score; based on the equipment adjustability score and combined with real-time production status data, a list of adjustable processes for the equipment is determined;

[0093] Candidate regulating equipment is determined based on the load forecast curve, equipment adjustability score and equipment adjustable process list. A multi-objective optimization function that includes electricity expenditure, production capacity loss and response default risk is constructed through a multi-objective optimization algorithm. Based on the candidate regulating equipment and the multi-objective optimization function, a load regulation strategy that includes equipment regulation objects, execution time period and load change is generated through the particle swarm optimization algorithm. Factory equipment is scheduled according to the load regulation strategy.

[0094] Example 3

[0095] A computer program product includes a computer program, which implements the steps of the method described in Example 2 when executed by a processor.

[0096] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. A load management system based on the load model of the plastic products industry, characterized in that: It includes: The load time series model construction module is used to obtain the equipment operating behavior based on the power curve, operating period and start-stop frequency of the equipment in each process of the factory process. The equipment is classified according to the equipment operating behavior, and the classification results and corresponding equipment operating behavior are annotated to obtain the load time series model. The data acquisition and fusion module is used to determine the monitoring points of the equipment in each process of the process according to the load time series model, collect the power, current, voltage and start-stop status data of the factory equipment in real time through the monitoring points to obtain real-time equipment operation data, and generate a real-time load data set in combination with the real-time production status data; The load forecasting and adjustability assessment module is used to train the historical operating load series of each plant equipment through a long short-term memory network based on the real-time load data set to generate a load forecast curve, use a support vector machine model to score the equipment adjustability based on the historical operating data of the plant equipment and generate an equipment adjustability score, and determine the equipment adjustable process list based on the equipment adjustability score and combined with real-time production status data; The scheduling execution module is used to determine candidate regulating equipment based on the load forecast curve, equipment adjustability score and equipment adjustable process list, and construct a multi-objective optimization function that includes electricity expenditure, production capacity loss and response default risk through a multi-objective optimization algorithm. Based on the candidate regulating equipment and the multi-objective optimization function, a load regulation strategy that includes equipment regulation objects, execution time period and load change is generated through the particle swarm optimization algorithm, and factory equipment is scheduled according to the load regulation strategy.

2. The load management system based on the plastics products industry load model according to claim 1, characterized in that: The types of factory equipment in the plastic products industry include twin-screw extruders, injection molding machines, internal mixers, coating rollers and foaming machines; the process flows include extrusion molding, injection molding, internal mixing and calendering, foaming molding and wet and dry methods; the equipment is classified according to its operating behavior to obtain the classification results as rigid load or adjustable load.

3. The load management system based on the plastics products industry load model according to claim 2, characterized in that: The specific process of constructing the load time series model is as follows: Obtain a list of equipment and process parameters corresponding to each process step in the plastics products industry; based on historical equipment operation data, extract the power change curve of each device during continuous operation and construct a time series of power consumption for each device; Based on the electricity consumption time series, the typical operating time periods, start-stop cycles and load peak intervals of each device are identified, the equipment operation continuity and operation volatility are determined to obtain the equipment operation characteristics, and the equipment is classified as rigid load or adjustable load according to the equipment operation continuity and operation volatility. The equipment is classified and labeled according to the equipment type, operation characteristics and process coupling relationship to obtain a load timing model.

4. The load management system based on the plastics products industry load model according to claim 1, characterized in that: The specific process of determining the monitoring points of the equipment in each process step of the process flow based on the load time series model, collecting the power, current, voltage and start / stop status data of the factory equipment in real time through the monitoring points to obtain real-time equipment operation data, and combining the real-time production status data to generate a real-time load data set is as follows: Based on the load time series model, monitoring points of equipment in each process step of the process flow are identified. Power monitoring devices and PLC control interfaces are deployed at the monitoring points of the twin-screw extruder, injection molding machine, internal mixer, coating roller, and foaming machine. During system operation, the power monitoring devices periodically collect power, current, voltage, and start / stop status data of each monitoring point to obtain real-time equipment operation data. The production schedule and process progress data are obtained through the production scheduling system. The start time, product batch, and operation section corresponding to each order are parsed using structured query language. The process plan start / stop time nodes and their deviation from the actual operation status are extracted to form production status tags to obtain real-time production status data. The real-time equipment operation data is standardized to eliminate abnormal spikes and short-term packet loss points. The Z-score algorithm is used to detect and mark outliers outside the range of three times the standard deviation. The real-time equipment operation data is smoothed by combining local weighted regression. Based on the segmented normalization strategy of the historical power distribution statistics of each type of equipment, the power, current and voltage are scaled according to the maximum value of the interval, and the start-stop status is encoded using 0-1. The power, current, voltage and start-stop status data of the equipment are associated with the production scheduling position and real-time status of each equipment in the production process. By combining the real-time equipment operation data and real-time production status data, a real-time load data set containing timestamp, equipment ID, process number, operation status, production scheduling progress and energy consumption indicators is constructed.

5. The load management system based on the plastics products industry load model according to claim 1, characterized in that: Based on the real-time load data set, the specific process of generating a load forecast curve by training the historical operating load sequence of each device in the factory through the long short-term memory network is as follows: The calculation formula of the long short-term memory network is as follows: h t =o t ⊙tanh(f t ⊙C t-1 +i t ⊙tanh(W c [x t ,h t-1 ,s t ]+b c )) Among them, h t is the load forecast result in the load forecast curve output by the long short-term memory network at time t; x t is the standardized load training sequence data at time t; h t-1 is the load forecast result of the load forecast curve output by the long short-term memory network at time t-1; C t-1 is the unit status at time t-1, which represents the accumulated information of the equipment operating status at the previous time t-1 and is used to predict the load change at the next time; t is the output value of the forget gate, which is used to control the unit state C at the previous moment t-1 The proportion of information that needs to be forgotten in t is the output value of the input gate, which is used to control the degree of update of the unit state by the current input information; t is the output value of the output gate, which is used to control the influence of the cell state on the current predicted output; W c The updated weight matrix for LSTM training is used to map device timing features to the update of unit state; b c The bias vector updated for LSTM training is used to adjust the prediction baseline of the LSTM model; t is the attention weight dynamically generated by the process feature attention layer according to the characteristics of different process stages at time t; ⊙ is the Hadamard product operation; The steps for training the historical operating load series of each device in the factory through the long short-term memory network in combination with the calculation formula of the long short-term memory network are as follows: Based on the power, current, voltage and start / stop status data of each device in the real-time load data set, combined with the real-time production status data of the device, the device load sequence is marked and divided according to the process stage, and the device load training data with process stage marks is generated; Normalizing the equipment load training data according to the load characteristic differences of each process to generate a standardized load training sequence; Constructing a long short-term memory network comprising an input layer, a long short-term memory layer, a process feature attention layer, and an output layer, taking the standardized load training sequence as input to the input layer, dynamically adjusting the output weight of the long short-term memory layer according to the characteristics of different process stages through the process feature attention layer, and generating a load forecast result by the output layer; The standardized load training sequence is input into the long short-term memory network for forward calculation. After obtaining the predicted output, the error between the load prediction result and the actual load is calculated. Based on the error, the weight parameters of each gate of the long short-term memory layer and the attention weight parameters of the process feature attention layer are updated through the back propagation algorithm until the prediction error converges to the preset threshold. Finally, the load prediction curve is generated by training.

6. The load management system based on the plastics products industry load model according to claim 1, characterized in that: The support vector machine model is used to score the equipment's adjustability based on the factory equipment's historical operating data and generate an equipment adjustability score. The specific process of determining the equipment's adjustable process list based on the equipment adjustability score and combined with real-time production status data is as follows: Based on the historical operation data of factory equipment, the load response records of each device within the preset time are extracted, and a feature vector containing the number of successful responses, the number of failed responses, the average response time, the advance notification time and the start-stop deviation rate is constructed. The support vector machine model is used for training. The support vector machine model uses the presence of successful response actions in the historical operation data of factory equipment as the classification standard. After the training is completed, the construction of nonlinear decision boundaries is realized through kernel function mapping; in the prediction stage, combined with the equipment start-stop state sequence in the real-time load data set, the current state vector is input into the target equipment, and the adjustability score of the target equipment within the specified response window is calculated. The scoring result is integrated with the key process plan in the real-time production status data, and logical constraints are used to determine whether there is adjustment space. The set of equipment with scores higher than the preset threshold and whose current operating status allows adjustment is screened out to determine the list of adjustable processes for the equipment.

7. The load management system based on the plastics products industry load model according to claim 1, characterized in that: The specific process of constructing a multi-objective optimization function that includes electricity expenditure, capacity loss, and response default risk through a multi-objective optimization algorithm is as follows: The formula for constructing a multi-objective optimization function including electricity expenditure, capacity loss and F response default risk through a multi-objective optimization algorithm is as follows: Wherein, F is the comprehensive target value of the multi-objective optimization function; T is the total number of time periods; P t is the electricity price information at time t; E t is the predicted power consumption at time t; C t is the unit capacity loss cost caused by equipment adjustment at time t; L t is the capacity loss at time t; R t D is the unit default cost of device response failure at time t; t is the probability of responding to default at time t; α is the weight coefficient corresponding to electricity expenditure, β is the weight coefficient corresponding to production capacity loss, and γ is the weight coefficient corresponding to the response default risk; The multi-objective optimization function is used to comprehensively evaluate the cost effects of different load regulation strategies in terms of electricity expenditure, capacity loss and response default risk; Indicates the sum of the product of unit electricity price and the corresponding power consumption in each forecast period, which is used to reflect the overall electricity expenditure; capacity loss Represents the product of the capacity loss caused by equipment participation in regulation and the unit capacity loss cost; responds to default risk It means estimating the default probability by the historical response reliability, and then quantifying the response default risk by multiplying it by the default penalty cost; the weight parameters α, β and γ corresponding to the above three items are configured according to demand and weighted aggregated.

8. The load management system based on the plastics products industry load model according to claim 1, characterized in that: Based on the candidate regulating equipment and the multi-objective optimization function, the particle swarm optimization algorithm is used to generate a load regulation strategy that includes the equipment regulation object, execution period, and load change. The specific process of scheduling factory equipment using the load regulation strategy is as follows: The formula of the particle swarm optimization algorithm is as follows: Among them, particle i represents a candidate solution of a device adjustment plan, and represents a set of specific device adjustment operations, including which devices, what method to use, and when to perform them; is the changing trend of the search direction of particle i in the velocity solution space at the k+1th iteration; is the changing trend of the search direction of particle i in the velocity solution space at the kth iteration; is the position of particle i at the kth iteration, which represents the combination of equipment start-stop or load reduction and the corresponding execution period; gbest is the individual historical best position of particle i up to the kth iteration, indicating the best solution position found in the history of particle i; k is the global optimal position of the particle swarm up to the kth iteration, indicating the optimal solution position known in the current entire particle swarm; ω is the inertia weight coefficient; c1 is the individual cognitive factor; c2 is the group cognitive factor; r1 is the first random number in the range of [0,1]; r2 is the second random number in the range of [0,1]; The particle swarm optimization algorithm is used to perform a global search for the adjustment combinations and adjustment periods of various equipment in the factory. The initial particle swarm is composed of multiple start-stop and load reduction combinations. Each particle represents a feasible adjustment strategy, and the particle position encoding includes equipment identification, operation mode and specific time window; for the initial particle swarm, particle swarm individuals corresponding to various equipment in the factory are generated according to the candidate adjustment equipment, and the executable time window is limited in combination with the process scheduling information. Based on the load forecast curve, the load change prediction value of the adjustment combination corresponding to each particle in the selected period is calculated, and the prediction value is substituted into the multi-objective optimization function for fitness calculation. In each round of iteration, the individual optimal position of the particle is obtained by comparing its current fitness with the historical optimal fitness. Comparison is updated, and the global optimal position is dynamically tracked according to the optimal fitness of the group; the speed vector is adjusted by combining the individual cognitive factor and the group cognitive factor, and the fusion guidance of the individual optimal and the global optimal is achieved through weighted summation to realize the update of the load regulation strategy. During the iterative process, when the overall fitness fluctuation of the particle swarm is lower than the set threshold or reaches the maximum number of iterations, it is considered to have converged, and the current global optimal load regulation strategy is output. In combination with the multi-objective optimization function, a load regulation strategy including the equipment regulation object, execution period and load change is selected, that is, the load regulation strategy with the lowest electricity price, the smallest production capacity loss and controllable response default risk, and the load regulation strategy is sent to the factory equipment for execution and scheduling through the PLC control interface.

9. A load management method based on a load model for the plastics products industry, characterized in that: It includes the following steps: The equipment operation behavior is obtained based on the power curve, operation period and start-stop frequency of the equipment in each process of the factory process. The equipment is classified according to the equipment operation behavior, and the classification results and the corresponding equipment operation behavior are labeled to obtain the load time series model; Determine the monitoring points of the equipment in each process step of the process according to the load time series model, collect the power, current, voltage and start-stop status data of the factory equipment in real time through the monitoring points to obtain real-time equipment operation data, and generate a real-time load data set in combination with the real-time production status data; Based on the real-time load data set, a long short-term memory network is used to train the historical operating load series of each device in the factory to generate a load forecast curve. A support vector machine model is used to score the equipment's adjustability based on the historical operating data of the factory equipment and generate an equipment adjustability score. Based on the equipment adjustability score and combined with real-time production status data, a list of adjustable processes for the equipment is determined; Candidate regulating equipment is determined based on the load forecast curve, equipment adjustability score and equipment adjustable process list. A multi-objective optimization function that includes electricity expenditure, production capacity loss and response default risk is constructed through a multi-objective optimization algorithm. Based on the candidate regulating equipment and the multi-objective optimization function, a load regulation strategy that includes equipment regulation objects, execution time period and load change is generated through the particle swarm optimization algorithm. Factory equipment is scheduled according to the load regulation strategy.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to claim 9 are implemented.

Images