Compound production control method and system based on neural network

Abstract

The invention discloses a compound production control method and system based on a neural network, and relates to the technical field of production control, and the method comprises the steps: inputting current process parameters related to the production quality of a compound into a pre-trained neural network model, and predicting the purity and yield of the compound produced based on the current process parameters; determining new process parameters of compound production according to the predicted purity and yield of the produced compound; the compound is produced according to the new process parameters, state information in a preset time period in the production process is obtained, and whether the production effect of the compound reaches the expectation or not is judged according to the state information; if the expectation is reached, compound production is continuously controlled based on the new process parameters; therefore, whether the production effect of the process parameters adjusted according to the prediction model in the actual compound production process reaches the expectation or not can be judged, the compound production is controlled in time, the loss of the compound production cost is reduced, and the production purity and yield are ensured to be qualified.

Description

Technical Field

[0001] The present invention relates to the technical field of production control, and in particular to a compound production control method and system based on neural network. Background Art

[0002] Neural network-based compound production control refers to the use of deep learning models to monitor and optimize key variables in the production process (such as temperature, pressure, reaction time, raw material concentration, etc.) in real time to improve the efficiency, stability and quality of compound synthesis; by training the neural network model, it can predict reaction results, identify abnormal situations, and automatically adjust process parameters, thereby achieving compound production control, reducing human intervention, and improving the consistency and economic benefits of compound production.

[0003] However, in actual neural network-based compound production control, some process parameters adjusted according to the prediction model may not achieve the expected production results in the actual compound production process, which may indicate a problem with the prediction model. If the process parameters are still adjusted based on the prediction model to control the production of the compound, it may lead to increased compound production costs and unqualified production purity and yield. Summary of the Invention

[0004] The purpose of the present invention is to solve the above-mentioned problems and provide a compound production control method and system based on neural network.

[0005] In a first aspect of the present invention, a compound production control method based on a neural network is first proposed, the method comprising: Inputting current process parameters related to compound production quality into a pre-trained neural network model to predict the purity and yield of the compound produced based on the current process parameters; determining new process parameters for compound production based on the predicted purity and yield of the compound produced; Produce compounds according to new process parameters, obtain status information during a preset time period during the production process, and determine whether the production effect of the compound meets expectations based on the status information; If the expectations are met, continue to control compound production based on the new process parameters.

[0006] In a second aspect of the present invention, a compound production control system based on a neural network is provided, the system comprising: Prediction module: Inputs the current process parameters related to compound production quality into a pre-trained neural network model to predict the purity and yield of the compound produced based on the current process parameters; Process parameter module: determines new process parameters for compound production based on the predicted purity and yield of the produced compound; Judgment module: produces compounds according to new process parameters, obtains status information within a preset time period during the production process, and judges whether the production effect of the compound meets expectations based on the status information; Control module: If the expected results are met, continue to control compound production based on the new process parameters.

[0007] Beneficial effects of the present invention: The present invention proposes a compound production control method and system based on a neural network, which predicts the purity and yield of the compound produced based on the current process parameters by inputting the current process parameters related to the quality of compound production into a pre-trained neural network model; determines new process parameters for compound production based on the predicted purity and yield of the produced compound; produces the compound based on the new process parameters, and obtains status information within a preset time period during the production process, and judges whether the production effect of the compound meets expectations based on the status information; if so, continues to control the compound production based on the new process parameters; in this way, during actual neural network-based compound production control, it can be judged whether the production effect of the process parameters adjusted according to the prediction model meets expectations in the actual compound production process, timely control the compound production, reduce the loss of compound production costs, and ensure that the purity and yield of the production are qualified. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The present invention will be further described below with reference to the accompanying drawings.

[0009] Figure 1 The figure is a flowchart of a compound production control method based on neural network; Figure 2 This is a framework diagram of a compound production control system based on a neural network. DETAILED DESCRIPTION

[0010] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0011] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0012] The embodiment of the present invention provides a compound production control method based on neural network. Figure 1 , Figure 1A flowchart of a neural network-based compound production control method provided in an embodiment of the present invention. The method includes the following steps: Inputting current process parameters related to compound production quality into a pre-trained neural network model to predict the purity and yield of the compound produced based on the current process parameters; determining new process parameters for compound production based on the predicted purity and yield of the compound produced; Produce compounds according to new process parameters, obtain status information during a preset time period during the production process, and determine whether the production effect of the compound meets expectations based on the status information; If the expectations are met, continue to control compound production based on the new process parameters.

[0013] Based on a neural network-based compound production control method provided by an embodiment of the present invention, through the above-mentioned method, during actual neural network-based compound production control, it is possible to judge whether the process parameters adjusted according to the prediction model have achieved the expected production effect in the actual compound production process, timely control the compound production, reduce the loss of compound production cost, and ensure that the production purity and yield are qualified.

[0014] In one embodiment, the steps of determining new process parameters for the current compound production based on the predicted purity and yield of the produced compound are: Recording the compound produced based on the current process parameters predicted by the pre-trained neural network model as a first purity and a first yield, respectively, and comparing the first purity and the first yield with a preset minimum purity and a preset minimum yield, respectively; If the first purity and the first yield are both not less than the preset minimum purity and the preset minimum yield, the current process parameters are used as new process parameters for compound production to produce the compound; If the first purity is less than the preset minimum purity, or the first yield is less than the preset minimum yield, new process parameters are calculated through the optimization algorithm until the predicted purity and yield of the produced compound are no less than the preset minimum purity and the preset minimum yield when predicted by the pre-trained neural network model based on the new process parameters.

[0015] It should be noted that the pre-trained neural network models mentioned above generally refer to deep learning models trained using extensive historical production data. Common models include feedforward neural networks (FNNs), long short-term memory networks (LSTMs), convolutional neural networks (CNNs), or deep reinforcement learning (DRL) models. The specific model chosen depends on the complexity of the compound production process, the data characteristics, and the prediction objectives. For example, a typical chemical synthesis reaction may involve multiple variables such as temperature, pressure, stirring speed, reaction time, and raw material ratios. These variables have a highly nonlinear relationship with the purity and yield of the final product, so deep feedforward neural networks (DNNs) are often used to model this mapping. Furthermore, if the production process exhibits time series characteristics, such as changes in certain parameters that affect the quality of subsequent products, LSTMs can be used to capture temporal dependencies and improve prediction accuracy. For complex image data (such as crystal morphology analysis under a microscope), CNNs can be used for feature extraction to assist in optimizing process parameters. For example, in the pharmaceutical industry, the synthesis of certain small molecule compounds requires monitoring the growth of crystal particles. Combining CNN and RNN for image analysis and time series prediction can more accurately adjust solvent concentration or crystallization temperature to achieve optimal purity and yield. Therefore, choosing the right neural network architecture is crucial to improving the intelligence level of compound production control.

[0016] It should be noted that the preset minimum purity and the preset minimum yield are set by professionals based on actual conditions and are not specifically limited or elaborated.

[0017] In one embodiment, the steps of calculating new process parameters by the optimization algorithm are: Initialize the process parameter range and set the upper and lower limits of the process parameters as the search space of the optimization algorithm; A set of initial process parameters is randomly generated from the set process parameter range through the particle swarm optimization algorithm, and each set of initial process parameters is used as an initial solution; The initial solution is input into the pre-trained neural network model to obtain the predicted compound purity and yield; The objective function value is calculated based on the predicted purity and yield. The objective function is calculated as follows: if the purity is less than the minimum purity, or the yield is less than the minimum yield, a penalty term is added. The calculated result is used as the objective function value of the corresponding solution. The optimization algorithm updates parameters: Based on the objective function value, the process parameters are adjusted to generate new candidate solutions, which are used as input for the next iteration.

[0018] The candidate solutions are iteratively adjusted until the particle swarm optimization algorithm finds a set of process parameters so that the predicted purity and yield both meet the minimum requirements, that is, the purity is not less than the minimum purity, and the yield is not less than the minimum yield. The process parameters of the corresponding solution are then used as the new process parameters.

[0019] It should be noted that this optimization method combines the particle swarm optimization (PSO) algorithm with a neural network prediction model, enabling efficient and accurate process parameter optimization within complex compound production control processes. Compared to traditional optimization methods based on empirical adjustments or simple regression models, this method offers several advantages: First, the PSO algorithm's global search capability allows it to efficiently explore multiple possible optimal solutions within a given process parameter range, avoiding local optima and ensuring the identification of more optimal process parameter combinations. Second, the neural network model is able to learn the complex relationships between nonlinear and high-dimensional variables in the compound production process, providing more accurate purity and yield predictions than traditional regression models, making the optimization process more reliable. Furthermore, this method employs a strategy that combines an objective function with a penalty function to ensure that the selected process parameters not only optimize production performance but also meet actual process constraints, avoiding unrealistic optimization results. This method can effectively reduce the number of trials and production costs, while simultaneously improving compound purity and yield, ensuring product quality meets industry standards. Therefore, this method not only advances the intelligentization of compound production but also significantly improves production efficiency and economic benefits in practical applications.

[0020] In one embodiment, the steps of determining whether the production effect of the compound has achieved the expected result based on the status information are as follows: The status information includes the purity yield abnormality coefficient, the energy consumption abnormality coefficient and the production time abnormality coefficient. The production effect abnormality coefficient is obtained based on the purity yield abnormality coefficient, the energy consumption abnormality coefficient and the production time abnormality coefficient, and the production effect abnormality coefficient is compared with the preset production effect abnormality coefficient threshold. Based on the comparison result, it is judged whether the production effect of the compound meets expectations.

[0021] In one embodiment, the steps for calculating the purity yield anomaly coefficient are: Obtain the number of compounds currently produced, detect the actual purity of each compound, and compare the actual purity of each compound with the preset minimum purity. If the actual purity is less than the preset minimum purity, the corresponding compound is marked as an abnormal compound; The purity abnormality quantity ratio is obtained by dividing the quantity of abnormal compounds by the quantity of compounds currently produced; Obtain the current actual yield, and divide the preset minimum yield by the actual yield to obtain the yield abnormality ratio; The purity-yield anomaly coefficient is obtained by adding the purity-yield anomaly ratio and the yield anomaly ratio.

[0022] It should be noted that the data acquisition methods involved in calculating the purity and yield anomaly coefficient primarily include real-time monitoring and sampling analysis during the production process. First, compound purity is typically determined through laboratory analytical techniques such as high-performance liquid chromatography (HPLC), gas chromatography (GC), and mass spectrometry (MS). Regular sampling and analysis of compound purity are performed to ensure compliance with established standards. Second, yield data is derived from the real-time recording and calculation of raw material inputs and product outputs during the production process. By accurately measuring raw material consumption and final product collection, the actual yield value can be determined. Other production process data, such as energy consumption and production time, are also monitored and collected in real time using sensors, flow meters, temperature control equipment, and other devices. All of this data can be centrally collected through industrial control systems (such as SCADA systems) and fed back to the optimization system in real time for subsequent anomaly coefficient calculation and production adjustments.

[0023] It should be noted that the Purity-Yield Abnormality Factor (PYA) is a comprehensive indicator that measures the degree to which the purity and yield of the actual product during the compound production process deviate from the preset standard (minimum requirement). It is composed of two factors: the Purity Abnormality Quantity Ratio (PUR) and the Yield Abnormality Ratio (YYA). The Purity Abnormality Quantity Ratio reflects the proportion of compounds with below-predetermined purity during the production process, while the Yield Abnormality Ratio measures the difference between the actual yield and the preset yield. By summing these two factors, the PYA reflects overall quality and yield issues during the production process. A higher PYA indicates a higher number of compounds that do not meet quality standards during the production process, potentially leading to a large number of substandard products. This not only indicates that the actual purity and yield of the compound do not meet expectations, but can also lead to increased production waste, such as excessive raw material and energy consumption, and may even impact subsequent production processes and increase costs. Furthermore, long-term production deviations can lead to increased equipment wear, maintenance costs, and downtime. Therefore, when a large purity yield abnormal coefficient occurs, it is necessary to stop production in time, adjust the process parameters or conduct system inspections to avoid further losses, ensure that the production line is restored to the predetermined optimized state, reduce the generation of defective products, and improve production efficiency and economic benefits.

[0024] In one embodiment, the calculation steps of the energy consumption abnormality coefficient are: Obtain real-time energy consumption data and the amount of compounds produced within a preset time period during the production process, design a time window to divide the real-time energy consumption data into several time windows, calculate the average energy consumption for each time window, and divide the average energy consumption by the amount of compounds produced in the corresponding time window to obtain the production energy consumption for the corresponding time window; Subtract the preset maximum energy consumption from the production energy consumption of each time window to obtain the energy consumption anomaly of the corresponding time window, and divide the energy consumption anomaly by the preset maximum energy consumption to obtain the energy consumption anomaly ratio; The energy consumption anomaly ratio of each time window is added together to obtain the energy consumption anomaly coefficient.

[0025] It should be noted that the data acquisition method involved in calculating the energy consumption anomaly coefficient primarily relies on real-time monitoring and measurement systems within the production process. Energy consumption data is typically collected in real time using energy monitoring instruments, flow meters, power meters, sensors, and other devices installed on production equipment. These instruments can measure key energy consumption indicators such as electricity consumption, steam usage, and fuel consumption, and are centrally managed through industrial control systems or data acquisition systems (such as SCADA systems). Energy consumption data for each time window can be segmented and statistically analyzed based on a set sampling frequency to obtain the average energy consumption for each time period. Simultaneously, compound quantities are recorded in real time through production management systems or quality inspection equipment to track production progress and product quantities. All of this data is transmitted to a central database via automated control systems for subsequent calculation and analysis, ensuring the accuracy and timeliness of energy consumption data, helping to optimize production processes and reduce energy waste.

[0026] It should be noted that the energy consumption anomaly coefficient is a comprehensive indicator that measures the deviation between actual energy consumption and preset standards during the production process. It calculates the difference between production energy consumption and the preset maximum energy consumption within each time window to determine the energy consumption anomaly ratio. The energy consumption anomaly ratios for each time window are then summed to obtain the final energy consumption anomaly coefficient. Specifically, the higher the production energy consumption and the greater the difference from the preset maximum energy consumption, the larger the energy consumption anomaly ratio, ultimately leading to an increase in the energy consumption anomaly coefficient. A larger energy consumption anomaly coefficient indicates that energy consumption during the production process far exceeds expectations, resulting in energy waste. Energy costs are one of the most important expenses in the production process. Excessive energy consumption not only increases production costs but may also indicate inefficient equipment, unreasonable process parameters, or equipment failure. For example, overloading equipment can lead to excessive energy consumption, while suboptimal production processes can also result in energy waste. Sustained high energy consumption not only affects economic efficiency but can also cause long-term damage to equipment, increase maintenance costs, or cause downtime. Therefore, when the energy consumption abnormality coefficient is large, it means that the production process fails to meet the expected energy efficiency standards. It is necessary to stop production in time, adjust process parameters or check equipment status to reduce energy waste, reduce production costs and avoid unnecessary losses.

[0027] In one embodiment, the calculation steps of the production time abnormality coefficient are: For each compound quantity produced within a preset time period during the production process, the actual production time of each compound at each production stage is obtained, and the actual production time of each production stage is compared with the preset standard production time. If the deviation between the actual production time and the standard production time for all production stages is within the preset range, the corresponding compound production time is qualified; If the deviation between the actual production time and the standard production time of any production stage is not within the preset range, the production time of the corresponding compound is unqualified; The total number of compounds with unqualified production time is divided by the number of compounds produced to obtain the production time abnormality coefficient.

[0028] It should be noted that the data used in calculating the production time anomaly coefficient is primarily obtained through real-time monitoring and recording systems within the production process. The actual production time for each compound at different stages is typically tracked and recorded in real time by the production equipment's control system and industrial control systems (such as SCADA). Using timers, sensors, and monitoring equipment installed throughout the production line, the system accurately records the time spent at each stage. Furthermore, standard production times are typically obtained from process documentation or production scheduling systems, based on preset time values based on process requirements, historical data, or optimized designs. Based on this, the system compares the actual production time for each compound with the preset standard time in real time and calculates the deviation to determine compliance. Furthermore, production quantity data can also be obtained from production management systems (such as ERP systems) to ensure the accuracy and timeliness of each batch's production data. All of this data, aggregated through automated systems, provides the foundational data for calculating the production time anomaly coefficient.

[0029] It should be noted that the production time deviation coefficient is a comprehensive indicator that measures the deviation between the actual production time of each compound and the preset standard production time. It reflects whether the actual production time of each stage in the compound production process meets the predetermined standard. If the actual production time of any production stage deviates from the standard time beyond the preset tolerance range, the production time of that compound is considered unsatisfactory, and the production time deviation coefficient is calculated. A larger production time deviation coefficient indicates that a large number of production links or compounds in the production process are completed within an unreasonable time frame. This may be caused by factors such as imperfect production processes, equipment failures, low operating efficiency, or improper process parameter settings. A large production time deviation coefficient indicates low production efficiency and the inability to complete production tasks on time. This not only wastes valuable production time, but also may affect the utilization of raw materials, lead to energy waste, and increase production costs. Prolonged production deviations may also lead to equipment overload, increase the risk of equipment failure, delay product delivery cycles, and cause customer dissatisfaction. Therefore, when the production time abnormality coefficient is high, it means that the production process has not met the expected efficiency requirements. It is necessary to stop production in time, find out the root cause of the problem and make adjustments to avoid further wasting time and resources, reduce production costs, reduce the generation of defective products, and ensure that production goals can be achieved on time.

[0030] In one embodiment, the steps of obtaining a production effect abnormality coefficient based on the purity yield abnormality coefficient, the energy consumption abnormality coefficient, and the production time abnormality coefficient, comparing the production effect abnormality coefficient with a preset production effect abnormality coefficient threshold, and determining whether the production effect of the compound meets expectations based on the comparison result are as follows: The purity yield abnormal coefficient, the energy consumption abnormal coefficient, and the production time abnormal coefficient are added together to obtain the production effect abnormal coefficient, and the production effect abnormal coefficient is compared with a preset production effect abnormal coefficient threshold. If the production effect abnormal coefficient is not less than the preset production effect abnormal coefficient threshold, it indicates that the production effect of the compound has not met expectations, and the production of the compound is stopped immediately. If the production effect abnormal coefficient is less than the preset production effect abnormal coefficient threshold, it means that the production effect of the compound has met expectations, and the compound production will continue to be controlled based on the new process parameters.

[0031] It should be noted that the preset production effect abnormality coefficient threshold is set by professionals based on actual conditions and is not limited or elaborated on in detail.

[0032] It should be noted that when a compound's production performance fails to meet expectations, this indicates that the current prediction model may contain issues or deviations, resulting in inaccurate predictions of purity, yield, energy consumption, or production time. Continuing to adjust process parameters based on an inaccurate prediction model can lead to a range of negative consequences. For example, the adjusted process parameters may fail to effectively improve production performance and may instead lead to problems such as waste of raw materials, increased energy consumption, and extended production times. This can increase production costs, reduce efficiency, and even produce substandard products, impacting product quality and market competitiveness.

[0033] Therefore, when the production performance anomaly coefficient exceeds the preset threshold, production must be halted immediately, the current prediction model must be suspended for parameter adjustment, and the model must be revalidated and optimized. The neural network model must be retrained or adjusted to incorporate more actual production data and feedback to improve the accuracy of the prediction model. Simultaneously, the production process must be carefully reviewed, including process flow, equipment status, and raw material quality, to ensure that production is not continuing with incorrect parameters. By promptly identifying and correcting deviations in the prediction model, unnecessary losses can be avoided and the production process can be restored to the expected quality and efficiency levels.

[0034] Based on the same inventive concept, the present invention also provides a compound production control system based on a neural network. Figure 2 , Figure 2 A framework diagram of a compound production control system based on a neural network provided in an embodiment of the present invention, the system includes: Prediction module: Inputs the current process parameters related to compound production quality into a pre-trained neural network model to predict the purity and yield of the compound produced based on the current process parameters; Process parameter module: determines new process parameters for compound production based on the predicted purity and yield of the produced compound; Judgment module: produces compounds according to new process parameters, obtains status information within a preset time period during the production process, and judges whether the production effect of the compound meets expectations based on the status information; Control module: If the expected results are met, continue to control compound production based on the new process parameters.

[0035] Based on a neural network-based compound production control system provided by an embodiment of the present invention, through the above-mentioned method, during actual neural network-based compound production control, it is possible to judge whether the process parameters adjusted according to the prediction model have achieved the expected production effect in the actual compound production process, timely control compound production, reduce the loss of compound production cost, and ensure that the production purity and yield are qualified.

[0036] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be used to artificially limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the application of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A compound production control method based on neural network, characterized in that: The following steps are involved: Inputting current process parameters related to compound production quality into a pre-trained neural network model to predict the purity and yield of the compound produced based on the current process parameters; determining new process parameters for compound production based on the predicted purity and yield of the compound produced; Produce compounds according to new process parameters, obtain status information during a preset time period during the production process, and determine whether the production effect of the compound meets expectations based on the status information; If the expectations are met, continue to control compound production based on the new process parameters.

2. The compound production control method based on neural network according to claim 1, characterized in that: The steps for determining new process parameters for the production of the current compound based on the predicted purity and yield of the produced compound are: Recording the compound produced based on the current process parameters predicted by the pre-trained neural network model as a first purity and a first yield, respectively, and comparing the first purity and the first yield with a preset minimum purity and a preset minimum yield, respectively; If the first purity and the first yield are both not less than the preset minimum purity and the preset minimum yield, the current process parameters are used as new process parameters for compound production to produce the compound; If the first purity is less than the preset minimum purity, or the first yield is less than the preset minimum yield, new process parameters are calculated through the optimization algorithm until the predicted purity and yield of the produced compound are no less than the preset minimum purity and the preset minimum yield when predicted by the pre-trained neural network model based on the new process parameters.

3. The compound production control method based on neural network according to claim 2, characterized in that: The steps for calculating new process parameters through the optimization algorithm are: Initialize the process parameter range and set the upper and lower limits of the process parameters as the search space of the optimization algorithm; A set of initial process parameters is randomly generated from the set process parameter range through the particle swarm optimization algorithm, and each set of initial process parameters is used as an initial solution; The initial solution is input into the pre-trained neural network model to obtain the predicted compound purity and yield; Calculate the objective function value based on the predicted purity and yield; The objective function is calculated as follows: if the purity is less than the minimum purity, or the yield is less than the minimum yield, a penalty term is added; the calculation result is used as the objective function value of the corresponding solution; The optimization algorithm updates parameters: Based on the objective function value, the process parameters are adjusted to generate new candidate solutions, which are used as input for the next iteration.

4. Iteratively adjust the candidate solutions until the particle swarm optimization algorithm finds a set of process parameters so that the predicted purity and yield both meet the minimum requirements, that is, the purity is not less than the minimum purity, and the yield is not less than the minimum yield. Then the process parameters of the corresponding solution are used as the new process parameters.

5. The compound production control method based on neural network according to claim 1, characterized in that: The steps to determine whether the production effect of the compound meets the expectations based on the status information are as follows: The status information includes a purity yield abnormality coefficient, an energy consumption abnormality coefficient, and a production time abnormality coefficient. The production effect abnormality coefficient is obtained based on the purity yield abnormality coefficient, the energy consumption abnormality coefficient, and the production time abnormality coefficient. The production effect abnormality coefficient is compared with a preset production effect abnormality coefficient threshold, and whether the production effect of the compound meets expectations is judged based on the comparison result.

6. The compound production control method based on neural network according to claim 4, characterized in that: The calculation steps of the purity yield anomaly coefficient are: Obtain the number of compounds currently produced, detect the actual purity of each compound, and compare the actual purity of each compound with the preset minimum purity. If the actual purity is less than the preset minimum purity, the corresponding compound is marked as an abnormal compound; The purity abnormality quantity ratio is obtained by dividing the quantity of abnormal compounds by the quantity of compounds currently produced; Obtain the current actual yield, and divide the preset minimum yield by the actual yield to obtain the yield abnormality ratio; The purity-yield anomaly coefficient is obtained by adding the purity-yield anomaly ratio and the yield anomaly ratio.

7. The compound production control method based on neural network according to claim 4, characterized in that: The calculation steps of the energy consumption abnormality coefficient are: Obtain real-time energy consumption data and the amount of compounds produced within a preset time period during the production process, design a time window to divide the real-time energy consumption data into several time windows, calculate the average energy consumption for each time window, and divide the average energy consumption by the amount of compounds produced in the corresponding time window to obtain the production energy consumption for the corresponding time window; Subtract the preset maximum energy consumption from the production energy consumption of each time window to obtain the energy consumption anomaly of the corresponding time window, and divide the energy consumption anomaly by the preset maximum energy consumption to obtain the energy consumption anomaly ratio; The energy consumption anomaly ratio of each time window is added together to obtain the energy consumption anomaly coefficient.

8. The compound production control method based on neural network according to claim 4, characterized in that: The calculation steps of the production time abnormality coefficient are: For each compound quantity produced within a preset time period during the production process, the actual production time of each compound at each production stage is obtained, and the actual production time of each production stage is compared with the preset standard production time. If the deviation between the actual production time and the standard production time for all production stages is within the preset range, the corresponding compound production time is qualified; If the deviation between the actual production time and the standard production time of any production stage is not within the preset range, the production time of the corresponding compound is unqualified; The total number of compounds with unqualified production time is divided by the number of compounds produced to obtain the production time abnormality coefficient.

9. The compound production control method based on neural network according to claim 4, characterized in that: The steps of obtaining a production effect abnormal coefficient according to the purity yield abnormal coefficient, the energy consumption abnormal coefficient and the production time abnormal coefficient, comparing the production effect abnormal coefficient with a preset production effect abnormal coefficient threshold, and judging whether the production effect of the compound reaches the expected level according to the comparison result are as follows: The purity yield abnormal coefficient, the energy consumption abnormal coefficient, and the production time abnormal coefficient are added together to obtain the production effect abnormal coefficient, and the production effect abnormal coefficient is compared with a preset production effect abnormal coefficient threshold. If the production effect abnormal coefficient is not less than the preset production effect abnormal coefficient threshold, it indicates that the production effect of the compound has not met expectations, and the production of the compound is stopped immediately. If the production effect abnormal coefficient is less than the preset production effect abnormal coefficient threshold, it means that the production effect of the compound has met expectations, and the compound production will continue to be controlled based on the new process parameters.

10. A compound production control system based on a neural network, used to implement the compound production control method based on a neural network according to any one of claims 1 to 8, characterized in that: The system comprises: Prediction module: Inputs the current process parameters related to compound production quality into a pre-trained neural network model to predict the purity and yield of the compound produced based on the current process parameters; Process parameter module: determines new process parameters for compound production based on the predicted purity and yield of the produced compound; Judgment module: produces compounds according to new process parameters, obtains status information within a preset time period during the production process, and judges whether the production effect of the compound meets expectations based on the status information; Control module: If the expected results are met, continue to control compound production based on the new process parameters.

Images