Vanadium electrolyte preparation threshold value regulation method and system based on emulsification probability prediction

Abstract

This invention belongs to the field of vanadium electrolyte preparation technology, specifically relating to a threshold control method and system for vanadium electrolyte preparation based on emulsification probability prediction. The method includes: real-time acquisition of motor electromagnetic torque and ultrasonic attenuation rate; mapping torque deviation levels and ultrasonic offset levels to obtain transient physical aging influence factors; obtaining a time-series baseline drift index based on the transient physical aging influence factors; obtaining a final fatigue ratio by weighted summation based on cumulative running time; multiplying the time-series baseline drift index by the final fatigue ratio to obtain an intermediate compensation requirement factor, obtaining a compensation intensity coefficient through a saturation mapping rule, and calculating a threshold compensation multiplier; dynamically reconstructing the decision threshold in a random forest model using the threshold compensation multiplier to predict the emulsification probability; issuing an early warning and control when the emulsification probability exceeds the threshold; and obtaining the vanadium electrolyte after control, static separation, and back-extraction. This invention improves the extraction efficiency of vanadium electrolyte.

Description

Technical Field

[0001] This invention relates to the field of vanadium electrolyte preparation technology, and in particular to a method and system for controlling the threshold of vanadium electrolyte preparation based on emulsification probability prediction. Background Technology

[0002] Vanadium electrolyte is the core material of vanadium redox flow batteries. Its preparation process usually involves solvent extraction to enrich and purify vanadium ions from vanadium-containing leachate. This process generally includes: thoroughly mixing the vanadium-containing aqueous phase with an organic extractant in an extraction tank to transfer vanadium ions from the aqueous phase to the organic phase; then allowing the phases to separate to obtain a vanadium-rich organic phase; and finally back-extracting the vanadium-rich organic phase to obtain the vanadium electrolyte. During the mixing process, if the mixing intensity is inappropriate or the properties of the two phases change, over-emulsification can easily occur. Once emulsification occurs, the oil-water separation interface disappears, which not only causes a sharp drop in vanadium recovery rate but also disrupts subsequent phase separation, back-extraction, and other processes. In severe cases, production needs to be stopped for cleaning, resulting in huge economic losses.

[0003] In actual industrial production, the extractant undergoes aging and degradation due to repeated use, and the agitator experiences mechanical wear during long-term operation, leading to over-emulsification during two-phase mixing. Existing technologies employ machine learning models such as random forests to predict emulsification risk, but these models typically use fixed thresholds for decision tree node partitioning. As the physical baseline shifts due to extractant aging and agitator wear, the threshold originally used for normal operating conditions becomes overly sensitive, causing frequent false alarms about emulsification or underreporting of actual emulsification events. In practical applications, this results in insufficient control stability, a lack of safety constraint mechanisms, and difficulty in ensuring the smoothness and safety of the threshold adjustment process. Furthermore, it lacks long-term adaptability to system characteristic shifts caused by gradual changes in operating conditions. Overall control robustness and reliability fail to meet the requirements of continuous industrial production, hindering accurate prediction of emulsification risk and stable control of the preparation process. This restricts the production efficiency and product consistency of vanadium electrolyte large-scale, continuous preparation. Summary of the Invention

[0004] To address the technical problems of emulsification causing production disruptions during the extraction process, the susceptibility of traditional machine learning models to false alarms and missed alarms due to fixed thresholds, the complexity and lack of safety constraints in adaptive threshold adjustment methods, and the difficulty in ensuring extraction stability and vanadium recovery rate due to the lack of coordinated optimization of multiple process parameters, this invention provides a threshold control method for vanadium electrolyte preparation based on emulsification probability prediction.

[0005] In a first aspect, this invention provides a threshold control method for vanadium electrolyte preparation based on emulsification probability prediction, comprising: real-time acquisition of the electromagnetic torque value of the stirring motor and the ultrasonic attenuation rate at the oil-water separation interface; determining the torque deviation level based on the deviation between the electromagnetic torque value and the reference torque value; determining the ultrasonic offset level based on the deviation between the ultrasonic attenuation rate and the reference attenuation rate, and obtaining a transient physical aging influence factor based on the mapping between the torque deviation level and the ultrasonic offset level; obtaining the time-series baseline drift index by performing time-sliding window weighted square accumulation and square root extraction on the transient physical aging influence factors of multiple consecutive batches; and calculating the extractant fatigue based on the cumulative running time of the extractant and the stirring impeller, respectively. The fatigue ratio and impeller fatigue ratio are calculated and weighted to obtain the final fatigue ratio. The time-series baseline drift index is multiplied by the final fatigue ratio to obtain the intermediate compensation demand factor. The intermediate compensation demand factor is then converted into a compensation intensity coefficient using the saturation mapping rule, and the threshold compensation multiplier is calculated based on the compensation intensity coefficient. The decision threshold of the decision tree nodes belonging to mechanical and fluid dynamic characteristics in the random forest model is dynamically reconstructed using the threshold compensation multiplier. The dynamically reconstructed random forest model is used to predict the emulsification probability of the current mixed phase. An early warning is given based on the comparison between the predicted emulsification probability and the threshold. The adjusted mixed phase is allowed to stand and separate, and then back-extracted to obtain vanadium electrolyte.

[0006] This method introduces dual physical quantity sensing based on motor electromagnetic torque and ultrasonic attenuation rate, and combines benchmark calibration and deviation level quantification to effectively construct a transient physical aging influencing factor that reflects extractant aging and impeller wear. This provides a reliable physical basis for subsequent adaptive threshold adjustment and solves the problem of false or missed emulsification reports caused by equipment baseline drift in the fixed threshold model.

[0007] Preferably, the step of mapping the transient physical aging influencing factor based on torque deviation level and ultrasonic offset level includes: firstly, mapping the torque deviation level... and ultrasonic offset level Square each element separately and add them together; then multiply the sum of squares by a scaling factor obtained by fitting a historical dataset using the least squares method. Round the product to the nearest integer; finally, compare the rounded value with 10 and take the smaller one as the transient physical aging influence factor. Among them, torque deviation level and ultrasonic offset level The values ​​of all values ​​are integers from 0 to 4, representing the transient physical aging influencing factor. Integers from 0 to 10.

[0008] This method maps transient physical aging influencing factors using an analytical formula based on the sum of squares operation, replacing the traditional fixed empirical value table. This allows the mapping rule to automatically determine the proportional coefficient by fitting historical data, adapting to the personalized aging characteristics of different extraction systems and equipment models, thus improving the model's generalization ability and transferability. At the same time, the sum of squares operation reflects the joint amplification effect of the deviation between the two physical quantities, and the rounding and upper limit constraints ensure the discreteness and safety of the output, reducing the online computational burden of industrial control systems.

[0009] Preferably, the step of obtaining the time-series baseline drift index by performing time-sliding window weighted summation and square root extraction on transient physical aging influencing factors from multiple consecutive batches includes: taking the most recent The transient physical aging impact factors of each batch are stored in a time sliding window. The transient physical aging impact factors of each batch are squared and multiplied by the exponential decay weight. The sum of these factors is accumulated to obtain a weighted sum of squares, and then the square root is taken to obtain the time series baseline drift index.

[0010] Preferably, the method for obtaining the intermediate compensation demand factor includes: calculating the ratio of the current cumulative operating time of the extractant to the upper limit of its rated lifespan to obtain the extractant fatigue ratio. The fatigue ratio of the agitator is obtained by calculating the ratio of its cumulative running time to its design life limit. Through weighted sum The final fatigue ratio was obtained. ,in The ratio of the time-series baseline drift index to the final fatigue degree is used to determine the relationship between the time-series baseline drift index and the final fatigue degree. Multiply to obtain the intermediate compensation demand factor .

[0011] This method integrates the fatigue ratio of extractant aging and impeller wear through a weighted summation method, which more accurately reflects the overall aging state of the equipment. By multiplying the time-series baseline drift index with the fatigue ratio, it achieves adaptive matching between the compensation intensity and the aging degree of the equipment throughout its entire life cycle, preventing false triggering in the early stages and providing reasonable compensation in the later stages.

[0012] Preferably, the step of converting the intermediate compensation demand factor into a compensation intensity coefficient through the saturation mapping rule includes: if the intermediate compensation demand factor ,but ;like ,but ;like ,but ;like ,but ;like ,but ; This is the compensation strength coefficient.

[0013] This method strictly limits the compensation intensity coefficient to between 0 and 0.95 through piecewise saturation mapping, ensuring that the threshold compensation multiplier does not exceed the safety limit. Even if the sensor fails completely or the data is extremely abnormal, the compensation multiplier will not go out of control. This provides a physical safety lock for the adaptive algorithm, ensuring the safety of industrial control.

[0014] Preferably, the step of calculating the threshold compensation multiplier based on the compensation intensity coefficient includes: In the formula, For threshold compensation multipliers; The maximum allowable threshold upward floating ratio, with a value range of [value range missing]. ; This is the compensation strength coefficient.

[0015] Preferably, the random forest model includes: using stirring speed, phase volume ratio, system temperature, pH value, and the transient physical aging influencing factor as input feature vectors; during model inference, the decision threshold of decision tree nodes belonging to mechanical and fluid dynamic features in the random forest model is dynamically reconstructed using the threshold compensation multiplier: if the feature of the current decision tree node belongs to the mechanical and fluid dynamic feature list, the static partitioning threshold of the node is multiplied by the threshold compensation multiplier to obtain a dynamic threshold, and the dynamic threshold is used for branch judgment; if it belongs to the thermodynamic and chemical feature list, the static threshold remains unchanged; the mechanical and fluid dynamic feature list includes stirring speed, phase volume ratio, and transient physical aging influencing factor, and the thermodynamic and chemical feature list includes system temperature and pH value.

[0016] This method performs targeted threshold reconstruction based on the physical properties of decision tree node features. It only compensates for the thresholds of features affected by mechanical aging, such as stirring speed, phase volume ratio, and transient physical aging influencing factors, while maintaining the original strict thresholds for chemically sensitive features, such as system temperature and pH value. This avoids the misjudgment of chemically sensitive dimensions caused by the traditional method of uniformly scaling the thresholds for all features, and improves the accuracy of emulsification probability prediction of the random forest model throughout its entire life cycle and the reliability of industrial deployment.

[0017] Preferably, the random forest model is trained by using historical production data, wherein the emulsification state is marked by manual observation or by the organic phase content in the aqueous phase exceeding 5% after phase separation.

[0018] Preferably, the early warning based on the comparison result of the predicted emulsification probability and the threshold includes: when the emulsification probability is greater than 75%, issuing a red alarm and executing at least one of the following control commands: reducing the stirring speed by 5% each time and not exceeding three times, adjusting the volume ratio of the organic phase to the aqueous phase from 1:1 to 0.8:1, or stopping stirring and allowing it to stand for 60 minutes to break the emulsion.

[0019] Secondly, the present invention provides a vanadium electrolyte preparation threshold control system based on emulsification probability prediction, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned vanadium electrolyte preparation threshold control method based on emulsification probability prediction is implemented.

[0020] By adopting the above technical solution, the above-mentioned threshold control method for vanadium electrolyte preparation based on emulsification probability prediction is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0021] The beneficial effects of this invention are as follows: By real-time acquisition of the electromagnetic torque value of the stirring motor and the ultrasonic attenuation rate of the oil-water separation interface, and by obtaining the transient physical aging influence factor based on benchmark calibration and deviation level quantification mapping, the influence of extractant aging and impeller wear on fluid dynamic characteristics is effectively quantified, providing a reliable physical basis for subsequent adaptive threshold adjustment and solving the problem of false or missed emulsification in the fixed threshold model due to equipment baseline drift; by performing time-sliding window weighted square accumulation and square root extraction on the transient physical aging influence factors of multiple consecutive batches to obtain the time-series baseline drift index, and combining the extractant fatigue ratio and impeller fatigue ratio to obtain the final fatigue ratio through weighted summation, the time-series baseline drift index is multiplied by the final fatigue ratio to obtain the intermediate compensation requirement factor, and then the intermediate compensation requirement factor is converted into a compensation intensity coefficient and the threshold compensation multiplier is calculated through piecewise saturation mapping rules. This mechanism not only achieves adaptive matching between the compensation intensity and the aging degree of the equipment throughout its entire life cycle, but also strictly limits the compensation intensity coefficient within a safe range through saturation mapping. Within the specified range, a physical safety lock is built into the adaptive algorithm to prevent compensation runaway caused by sensor anomalies or extreme data. During the inference process of the random forest model, threshold compensation multipliers are used to dynamically reconstruct the decision thresholds of decision tree nodes belonging to mechanical and fluid dynamic features in the random forest model. That is, only the stirring speed, phase volume ratio, and transient physical aging influencing factors in the mechanical and fluid dynamic feature list are compensated for with thresholds, while the system temperature and pH value in the thermodynamic and chemical feature list are kept at static thresholds. This directional threshold reconstruction strategy avoids the misjudgment of chemically sensitive dimensions caused by the traditional method of uniformly scaling the thresholds of all features. It realizes the synergistic optimization of multiple process parameters and significantly improves the accuracy of emulsification probability prediction of the random forest model throughout the entire life cycle. Finally, based on the comparison results of the predicted emulsification probability and the threshold, graded early warning is performed and corresponding control commands are executed. The adjusted mixed phase is then allowed to stand for phase separation and back-extracted to obtain vanadium electrolyte, which effectively improves the stability of the extraction process and the vanadium recovery rate, and finally obtains a high-quality vanadium electrolyte product. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the process framework of the vanadium electrolyte preparation threshold control method based on emulsification probability prediction in this invention. Figure 2 This is a schematic diagram showing the distribution of ultrasonic attenuation rate under normal operating conditions; Figure 3 This is a schematic diagram showing the distribution of ultrasonic attenuation rate under moderate emulsification conditions; Figure 4 This is a schematic diagram illustrating the distribution of ultrasonic attenuation rate under severe emulsification conditions; Figure 5 This is a schematic diagram showing a comparison of the detection effects of the method of the present invention and the conventional method. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] This invention discloses a threshold control method for vanadium electrolyte preparation based on emulsification probability prediction, referring to... Figure 1 This includes steps S1 to S5: S1. Obtain comprehensive characteristic data of the extraction process based on online sensing of multi-source physical quantities and calibration of operating conditions.

[0026] It should be noted that this step aims to construct a comprehensive feature system that can reflect extractant aging, impeller wear, and two-phase emulsification state through online sensing of multi-dimensional physical quantities and calibration of operating conditions. This provides a stable and reliable data foundation for subsequent extraction process parameter prediction and emulsification identification, and avoids distortion of the prediction model due to operating condition baseline drift.

[0027] Specifically, a torque acquisition module is connected in parallel to the output of the frequency converter of the stirring motor in the vanadium electrolyte extraction tank. The electromagnetic torque value of the motor is acquired in real time at a sampling frequency of 10 times / second. The electromagnetic torque value of the motor is used as a real-time mechanical characteristic. The electromagnetic torque value of the motor is positively correlated with the fluid resistance of the stirring paddle. An ultrasonic pulse echo level gauge is installed at the oil-water separation interface of the extraction tank. Ultrasonic waves are emitted to the phase interface at a fixed period of 5 seconds / wave and the echo signal is received. The difference between the amplitude of the emitted ultrasonic wave and the amplitude of the echo signal is calculated. The ratio of this difference to the amplitude of the emitted ultrasonic wave is used as the energy attenuation rate of the ultrasonic wave after propagation through the mixed phase. The energy attenuation rate is used to characterize the degree of formation of micro-emulsion droplets and belongs to the real-time interface characteristics.

[0028] Furthermore, the system automatically reads the factory operation and maintenance log database through the interface to obtain runtime data for each batch, specifically including: the cumulative runtime of the current extractant batch, the rated lifespan of the extractant model under standard operating conditions, and the cumulative runtime of the agitator since the last replacement. The above runtime data is stored in the real-time data table of the host computer. The real-time data table of the host computer refers to a relational data structure deployed in the memory of the host computer, which is used to store key process parameters and equipment status variables of each batch by timestamp, and supports fast query and update.

[0029] Specifically, before the equipment is installed, debugged, and put into operation for the first time, the system performs a benchmark calibration process: standard clean organic phase and aqueous phase are injected into the extraction tank, and the mixture is stirred stably at the rated speed for 30 minutes. The electromagnetic torque value of the motor collected at this time is calibrated as the benchmark mechanical characteristic, and the ultrasonic energy attenuation rate at the corresponding moment is calibrated as the benchmark interface characteristic. The benchmark mechanical characteristic and benchmark interface characteristic are encrypted using the AES-256 symmetric encryption algorithm and stored in the non-volatile memory of the host computer. The benchmark calibration operation can only be re-performed by authorized personnel through a dedicated maintenance terminal when core components of the equipment, such as the stirring paddle or ultrasonic probe, are replaced. All sensing data is uploaded to the central host computer real-time database in real time via industrial Ethernet at a frequency of 1 time / second. The sensing data includes real-time mechanical characteristics, real-time interface characteristics, benchmark mechanical characteristics, benchmark interface characteristics, and runtime data for each operation. The host computer adopts a dual-machine hot standby redundant server architecture to ensure the reliability of data transmission and storage.

[0030] Please see Figure 2 , Figure 2 This diagram schematically illustrates the distribution of ultrasonic attenuation rate under normal operating conditions. As shown in the diagram, the time-domain signal of the ultrasonic attenuation rate under normal extraction conditions is generally stable without drastic fluctuations, exhibiting a typical normal distribution. The mean ultrasonic attenuation rate is... The normal fitting parameters are , The data distribution is concentrated and has a small degree of dispersion, fluctuating only slightly around the mean, with no obvious deviation or abnormal peaks. This directly reflects that the oil and water phases in the extraction system are mixed evenly, no micro-emulsion droplets are generated, the interface state is stable, and the emulsification probability is extremely low, which is consistent with the ideal operating state of the vanadium electrolyte preparation and extraction process.

[0031] S2. Obtain the transient physical aging influence factor based on the mapping between torque deviation level and ultrasonic offset level.

[0032] It should be noted that this step aims to quantify the degree of deviation between real-time mechanical features and reference mechanical features, and the degree of offset between real-time interface features and reference interface features, respectively. Then, using a pre-constructed analytical formula based on the sum of squares operation, the transient physical aging influence factor of the current batch is mapped, thereby quantifying the instantaneous impact of extractant aging and impeller mechanical wear on fluid dynamics, and providing a basic input for subsequent time series trend analysis.

[0033] Specifically, the system first calculates the difference between the motor's electromagnetic torque value and the reference torque value, divides this difference by the reference torque value to obtain the torque deviation ratio, and then classifies the torque deviation level according to the absolute value of the torque deviation ratio: Let the motor's electromagnetic torque value be... The reference torque value is The torque deviation ratio is calculated as follows: The torque deviation level is then determined based on the absolute value of this ratio; the threshold parameter used for this determination is... And satisfy These parameters are determined by the following method: the collection device is operating stably in a brand new state. batch Calculate the mean of the absolute value of the torque deviation ratio data. and standard deviation Then set , , , The specific division rules are as follows: when the absolute value of the torque deviation ratio is less than... When the absolute value of the torque deviation ratio is greater than or equal to a certain level, it is defined as level 0, indicating no deviation; when the absolute value of the torque deviation ratio is greater than or equal to a certain level, it is defined as level 0, indicating no deviation. and less than When the absolute value of the torque deviation ratio is greater than or equal to level 1, it indicates a slight deviation; when the absolute value of the torque deviation ratio is greater than or equal to level 1, it indicates a slight deviation. and less than When the absolute value of the torque deviation ratio is greater than or equal to a certain level, it is defined as Level 2, indicating a significant deviation; when the absolute value of the torque deviation ratio is greater than or equal to a certain level. and less than When the absolute value of the torque deviation ratio is greater than or equal to level 3, it indicates a severe deviation; When the deviation is defined as level 4, it indicates an extremely severe deviation. The system also records the direction of the torque deviation and generates a direction flag: if the motor electromagnetic torque value is greater than the reference torque value, the direction flag is positive, indicating that the stirring resistance has increased; if the motor electromagnetic torque value is less than the reference torque value, the direction flag is negative, indicating that the stirring resistance has decreased.

[0034] It should be noted that the grading depends only on the absolute value of the deviation ratio. Therefore, regardless of whether the direction flag is positive or negative, a uniform grading value is determined according to the above threshold, i.e., grading from 0 to 4. The direction flag does not participate in the grading, but it will be retained for subsequent judgment. That is, when the torque deviation grading reaches 2 or above, the direction flag is negative, and the ultrasonic offset grading reaches 1 or above, the system will forcibly set the transient physical aging influence factor to 9.0 and send a prompt message to the host computer: "Abnormal negative torque drift detected, which may be caused by the following reasons: mechanical failure of the agitator, excessive dilution of the extractant, decrease in material viscosity, or agitator detachment, etc. Please check the equipment and process parameters in time."

[0035] Furthermore, the system also calculates the real-time ultrasonic attenuation rate. Compared with the reference ultrasonic attenuation rate The deviation ratio, i.e. The threshold parameter used for the division is And satisfy The method for determining these parameters is the same as the method for determining the threshold parameters of the torque deviation level, that is, based on the statistical distribution of historical ultrasonic attenuation rate deviation ratio data during brand-new and stable operation of the equipment, data on stable operation of the equipment in a brand-new state are collected. batch The mean of the absolute values ​​of the ultrasonic attenuation rate deviation ratio data and standard deviation Then set , , , The classification rules for ultrasonic deviation levels are as follows: when the absolute value of the deviation ratio is less than... When the absolute value of the deviation ratio is greater than or equal to a certain value, it is defined as level 0, indicating no offset; when the absolute value of the deviation ratio is greater than or equal to a certain value, it is defined as level 0, indicating no offset. and less than When the absolute value of the deviation ratio is greater than or equal to level 1, it indicates a slight deviation; when the absolute value of the deviation ratio is greater than or equal to... and less than When the absolute value of the deviation ratio is greater than or equal to Level 2, it indicates a significant deviation; when the absolute value of the deviation ratio is greater than or equal to... and less than When the absolute value of the deviation ratio is greater than or equal to Level 3, it indicates a severe deviation; At this point, it is defined as level 4, indicating an extremely severe offset.

[0036] It should be noted that the positive and negative values ​​of the ultrasonic attenuation rate deviation ratio have clear physical meanings: when the deviation ratio is positive, it indicates that the real-time ultrasonic attenuation rate is greater than the reference value, that is, the ultrasonic energy attenuation is increased, which indicates that the degree of formation of micro-emulsion droplets is intensified and the emulsification tendency is enhanced; when the deviation ratio is negative, it indicates that the real-time ultrasonic attenuation rate is less than the reference value, that is, the ultrasonic energy attenuation is reduced, which indicates that the two phases are too clear and the risk of emulsification is low.

[0037] Please see Figure 3 , Figure 3 The diagram schematically illustrates the distribution of ultrasonic attenuation rate under moderate emulsification conditions. As can be seen from the diagram, the mean ultrasonic attenuation rate is approximately 35.1%, and the standard deviation is approximately 4.5%, which is significantly higher than the mean of 12.0% under normal operating conditions. Furthermore, the data dispersion is increased, indicating that a certain degree of micro-emulsified droplets have appeared, but emulsification has not yet been complete.

[0038] Please see Figure 4 , Figure 4 The diagram illustrates the distribution of ultrasonic attenuation rate under severe emulsification conditions. As can be seen from the diagram, the average ultrasonic attenuation rate is as high as 78.5%, with very little fluctuation. This indicates that a large number of stable emulsion droplets have formed at the interface between the two phases, and the ultrasonic energy is rapidly attenuated. At this time, a severe emulsification accident in which oil and water cannot be separated is very likely to occur.

[0039] Specifically, the system determines the transient physical aging influencing factor based on the following relationship, denoted as: Transient physical aging influencing factors It is obtained through a predefined mapping relationship, which is constructed as follows: First, the torque deviation level is... and ultrasonic offset level Square each element separately and add them together; then multiply the sum of squares by a scaling factor obtained by fitting a historical dataset using the least squares method. Round the product to the nearest integer; finally, compare the rounded value with 10 and take the smaller one as the transient physical aging influence factor. Among them, torque deviation level and ultrasonic offset level The values ​​of all values ​​are integers from 0 to 4, representing the transient physical aging influencing factor. Integers from 0 to 10.

[0040] Furthermore, the historical dataset contains multiple sets of torque deviation levels, ultrasonic offset levels, and their corresponding quantitative values ​​of actual physical aging; in a specific instance of this embodiment, the fitted... See Table 1: Table 1

[0041] Note: This table is only an example constructed based on predefined mapping relationships. In actual applications, the scaling factor... The data needs to be refitted based on the specific process data, and the contents of the table will change accordingly.

[0042] In addition, if the following conditions are met simultaneously: torque deviation level reaches level 2 or above, torque deviation direction is negative, and ultrasonic offset level reaches level 1 or above, then a forced setting will be implemented. The priority of this mandatory rule is higher than the calculation result of the above relation. When this mandatory rule is triggered, the system sends a prompt message to the host computer: "Abnormal negative torque drift detected, which may be caused by the following reasons: mechanical failure of the agitator, excessive dilution of the extractant, decrease in material viscosity, or agitator detachment, etc. Please check the equipment and process parameters in time." In the state of brand-new equipment, the torque deviation level and ultrasonic offset level are both 0. Substituting into the formula, we get... The deviation level increases as the equipment ages. The increase according to the above rules reflects the positive correlation between the degree of aging and the influencing factors.

[0043] S3. Obtain the time-series baseline drift index based on the time-sliding window weighted square accumulation and the forgetting coefficient.

[0044] It should be noted that this step aims to eliminate the influence of random disturbances in a single batch by introducing a dynamic time sliding window and an exponential decay forgetting mechanism to perform weighted accumulation processing on the transient physical aging influencing factors of multiple consecutive batches, thereby extracting the time-series baseline drift index that reflects the long-term aging trend of the equipment, and providing stable trend characteristics for subsequent threshold compensation.

[0045] Specifically, the system allocates a circular buffer in the host computer's memory to store the most recently accessed data. Transient physical aging influencing factors of each batch In this embodiment The value is set to 50, which can be adjusted according to the actual production batch frequency. After each batch extraction operation is completed, the system pushes the transient physical aging impact factor of the current batch into the sliding window and automatically discards the oldest batch data in the sliding window, ensuring that the most recent batch data is always stored in the sliding window. Data from each batch.

[0046] Furthermore, the system assigns a weight to each batch within the sliding window, following a time priority rule: the newest batch has the highest weight, and this highest weight is recorded as... Exemplary When backtracking from the latest batch to the earliest batch, the weight is multiplied by a fixed forgetting coefficient for each previous batch. In this embodiment The value is 0.85; the sequence number of the batches within the sliding window, from oldest to newest, is recorded as follows. ,in Representing the latest batch, then the number The weight of the batch is For example, when At that time, the latest batch, namely the [number]th The batch weight is 1.0, the first The batch weight is 0.85 to the power of 1, which is 0.85. The batch weight is 0.85 to the power of 2, which is 0.7225, and so on.

[0047] Specifically, the system squares the transient physical aging impact factor for each batch within the sliding window, obtaining the square of the transient physical aging impact factor for each batch. The purpose of the squaring operation is to reduce the impact of small fluctuations on subsequent cumulative results, while significantly amplifying the impact of large outliers, thereby improving the sensitivity to severe aging events. The system starts from the oldest batch in the sliding window, i.e. From the beginning of the batch to the latest batch, i.e. After each batch is completed, the square of the transient physical aging impact factor for each batch is multiplied by the weight corresponding to that batch. Then, all weighted results are summed to obtain a weighted sum of squares. Finally, for the weighted sum of squares Performing the square root operation yields the time series baseline drift index, denoted as . Time-series baseline drift index Reflects the equipment in the past The cumulative effect of aging and wear in each batch, the closer the outlier is to the current batch, the better. The greater the contribution of older data, the more exponentially the contribution decreases; if a batch shows an extremely high transient physical aging impact factor but subsequent batches return to normal, due to the forgetting factor, this outlier will affect the time series baseline drift exponentially after several batches. The effect gradually weakens and will not permanently increase the drift index.

[0048] S4. Obtain the threshold compensation multiplier based on the fatigue ratio and saturation mapping rule.

[0049] It should be noted that this step aims to combine the time-series baseline drift index with the remaining life of the equipment by introducing the fatigue ratio of the extractant and the agitator, and then limit the compensation requirement to a preset safety limit through the saturation mapping rule, thereby generating a threshold compensation multiplier for adjusting the decision boundary of the prediction model, ensuring that the adaptive process is always under control and does not exceed the engineering safety boundary.

[0050] Specifically, the system first calculates the cumulative runtime of the current extractant. Its rated life limit The ratio of the two values ​​is used to obtain the extractant fatigue ratio. ,Right now Similarly, if the cumulative runtime of the agitator since the last replacement is recorded... and its design life limit Similarly, the fatigue ratio of the agitator is calculated. Final fatigue ratio Calculated using a weighted sum method, the relation is: ,in, For the weighting coefficients, satisfying The weighting coefficients reflect the difference in the contribution of extractant aging and impeller wear to the system's emulsification risk. These coefficients can be determined through historical data regression analysis or expert experience. In this embodiment, based on fitting of a large amount of experimental data, extractant aging has a greater impact on baseline drift; therefore, the weighting coefficient is chosen as the weighting factor. , , The range of values ​​is clamped to Within the range.

[0051] Furthermore, the system will use the time-series baseline drift index. Ratio to final fatigue Multiply to obtain the intermediate compensation demand factor. ,Right now Intermediate compensation demand factor The physical meaning is: the time-series baseline drift index This reflects the severity of the equipment's current deviation from its ideal state, and ultimately, the fatigue ratio. This reflects the persistence of this deviation, namely the percentage of the equipment's service life. The product of the two means that even in the early stages of equipment operation, deviations caused by accidental factors can lead to further deviations. Too high, but due to Very small Still low; and even later in the equipment's lifespan, At a moderate level Also because The larger the scale, the higher the demand for compensation.

[0052] Specifically, to prevent the compensation multiplier from increasing indefinitely, the system employs a saturation mapping rule to reduce the intermediate compensation demand factor. Mapped to compensation strength coefficient , The value range is 0 to 0.95; the thresholds and coefficients of the following saturation mapping rules are exemplary values ​​and can be adjusted according to actual process safety requirements: If ,but ;like ,but ;like ,but ;like ,but ;like ,but Upon reaching the saturation upper limit, the above piecewise linear mapping results in the compensation intensity coefficient... Follow It increases monotonically, but never exceeds 0.95.

[0053] Furthermore, the system presets a maximum allowable threshold upward fluctuation ratio. This value is set by process experts based on the tolerance of the extraction system, representing the maximum allowable threshold fluctuation. The range of values ​​is In this embodiment The preferred value is 0.15, indicating that the threshold is allowed to float up by a maximum of 15%, then the threshold compensation multiplier... The calculation formula is: For example, when and hour, This indicates that the model's decision threshold has been increased by 7.5%; when When the saturation value of 0.95 is reached, It shall not exceed the preset upper limit of 1.15.

[0054] It should be noted that even if the intermediate compensation demand factor is reduced due to complete sensor failure... The calculation is infinity, and the saturation mapping rule will also affect the compensation intensity coefficient. The clamp is set at 0.95 to ensure the compensation multiplier. Never exceed This mechanism is equivalent to building a physical security lock into the adaptive algorithm, ensuring the security of industrial control.

[0055] S5. Obtain emulsification probability and early warning instructions by reconstructing based on characteristic physical properties and dynamic thresholds.

[0056] It should be noted that this step aims to use a pre-trained random forest model, combined with the generated threshold compensation multiplier, to perform directional threshold reconstruction on the extraction process parameters. That is, the decision boundary is dynamically adjusted only for features affected by mechanical aging, while the original threshold is maintained for chemically sensitive features. Finally, the emulsification probability is output and the corresponding warning instructions are executed according to the probability level.

[0057] Specifically, in historical production data, emulsification status is marked through manual observation or subsequent separation effects. For example, emulsification is determined by detecting that the organic phase content entrained in the aqueous phase after phase separation exceeds 5%. Then, the system trains a random forest classification model using historical production data in the offline phase. In this embodiment, the number of training samples is at least 200 batches, including at least 100 emulsified samples and at least 100 non-emulsified samples to ensure class balance. The hyperparameters of the random forest are set as follows: at least 100 decision trees, a maximum depth of at least 10 layers per tree, and the number of features randomly selected when splitting at each node is the square root of the total number of input features (i.e., 5 input features, randomly selected each time). Two options are selected, with a minimum sample size of 5 for each leaf node. It should be noted that the above values ​​are merely exemplary lower limits. In practical applications, using more training samples, such as 2000 batches, and larger tree models, such as 500 decision trees, can achieve more stable prediction performance, but this invention is not limited to these. The input feature vector of the random forest classification model includes stirring speed, phase volume ratio, system temperature, pH value, and transient physical aging influencing factors. The model output is the emulsification probability, ranging from 0% to 100%. After training, the system extracts the feature name and its partitioning threshold corresponding to each internal node of each decision tree in the model, and records these original thresholds as static partitioning thresholds. .

[0058] Furthermore, when the system is running online, after a new batch of data enters the model prediction process, it begins to traverse each decision tree in the random forest. Upon reaching each internal node, the system first reads the feature name used by that node and then determines the category to which the feature belongs. The system pre-establishes two feature lists: the mechanical and fluid dynamics feature list includes stirring speed, phase volume ratio, and transient physical aging influencing factors, which are directly affected by impeller wear, extractant aging, and changes in fluid resistance; the thermodynamic and chemical feature list includes system temperature and pH value, which are mainly determined by chemical reaction equilibrium and thermodynamic conditions and are not affected by mechanical aging.

[0059] Specifically, the system performs targeted threshold reconstruction based on feature categories. If the current node's features belong to the mechanical and fluid dynamics feature list, then the static partitioning threshold for that node is retrieved. Combine it with the threshold compensation multiplier Multiply to obtain the dynamic threshold and use The boundary condition for this node is defined as follows: when the value of this feature in the input data is greater than or equal to... If the current node's feature belongs to the thermodynamic and chemical feature list, then the compensation multiplier is forcibly set to 1, i.e., the dynamic threshold. Without making any adjustments to the threshold, it is important to note that during the model inference process, for each decision tree node, the system calculates the dynamic threshold in real time and replaces the original static threshold for branch judgment.

[0060] It should be noted that the calculation of the transient physical aging impact factor does not depend on the output of the random forest model, so there is no circular dependency. The transient physical aging impact factor serves as both the model input feature and the source of the threshold compensation multiplier, enabling the model to dynamically adjust its sensitivity to the transient physical aging impact factor according to the current aging level. This is one of the core innovations of this invention.

[0061] Specifically, for comparison, the traditional fixed threshold is set at 65%, please refer to [link to relevant documentation]. Figure 5 , Figure 5 The diagram illustrates a comparison between the detection performance of the dynamic threshold of this invention and the traditional fixed threshold. As can be seen from the diagram, with the increase of operating time, equipment aging leads to physical baseline drift. The traditional fixed threshold (65%) shows large-scale detection failure in the later stages, and cannot identify the actual emulsification risk. In contrast, this invention adopts a dynamic threshold compensation strategy based on the time-series baseline drift index and fatigue ratio. The threshold can adaptively rise, always maintaining effective detection of emulsification risk, and significantly extending the applicability of the model throughout its entire life cycle.

[0062] Furthermore, after all decision trees complete their branch paths according to the dynamically reconstructed thresholds, each tree outputs an emulsification probability value, ranging from 0 to 1. The random forest then performs an arithmetic average of the output probabilities of all trees to obtain the final emulsification probability. The system is based on the emulsification probability. The numerical range determines the level of response: if The system displays "normal" on the user interface and does not issue any intervention commands; if The system issues a yellow warning signal, prompting operators to observe the oil-water separation interface in the extraction tank; if The system issues a red alarm signal and automatically sends control commands to the distributed control system. These commands include, but are not limited to: reducing the stirring speed, for example, by 5% each time, not exceeding three times; adjusting the feed flow ratio to reduce the proportion of the organic phase, for example, adjusting the ratio from 1:1 to 0.8:1; or pausing the current batch and entering the demulsification cycle mode, i.e., stopping stirring and extending the settling time, and recording it in the system event log for subsequent analysis and process optimization. Here, 50% and 75% are exemplary values ​​and can be adjusted according to actual needs. Subsequently, the adjusted or confirmed safe mixed phase is allowed to stand for phase separation, and the obtained vanadium-rich organic phase is subjected to conventional back-extraction to obtain the vanadium electrolyte product.

[0063] This invention also provides a vanadium electrolyte extraction emulsification risk prediction system for implementing the above method. The system includes a sensing layer, a transmission layer, a computation and control layer, and an execution layer. The sensing layer includes a torque sensor integrated into the frequency converter, an ultrasonic level gauge, and a maintenance log reading interface. The transmission layer includes an industrial Ethernet switch and a fiber optic link. The computation and control layer includes an industrial host computer and a distributed control system controller, wherein the industrial host computer is equipped with a real-time database, a rule engine, and a random forest inference engine. The execution layer includes a stirring motor frequency converter and a feed regulating valve. The host computer stores baseline calibration values, the proportional coefficient of transient aging influence factors, sliding window forgetting weight configurations, a saturation mapping rule table, a feature classification library, a trained random forest model, and its static partitioning threshold. The host computer periodically executes the above steps, updating the sliding window after each batch.

Claims

1. A threshold control method for vanadium electrolyte preparation based on emulsification probability prediction, characterized in that, include: Real-time acquisition of the electromagnetic torque value of the stirring motor and the ultrasonic attenuation rate at the oil-water separation interface; The torque deviation level is determined based on the deviation between the motor's electromagnetic torque value and the reference torque value; the ultrasonic offset level is determined based on the deviation between the ultrasonic attenuation rate and the reference attenuation rate; and the transient physical aging influence factor is obtained by mapping the torque deviation level and the ultrasonic offset level. The time-series baseline drift index is obtained by weighted summation and square rooting of the transient physical aging influencing factors from multiple consecutive batches using a sliding window. The fatigue ratios of the extractant and impeller are calculated based on their cumulative running time, and the final fatigue ratio is obtained by weighted summation. The intermediate compensation demand factor is obtained by multiplying the time-series baseline drift index by the final fatigue ratio. The intermediate compensation demand factor is then converted into a compensation intensity coefficient using a saturation mapping rule, and the threshold compensation multiplier is calculated based on the compensation intensity coefficient. The decision thresholds of decision tree nodes belonging to mechanical and fluid dynamic characteristics in the random forest model are dynamically reconstructed using threshold compensation multipliers. The dynamically reconstructed random forest model is used to predict the emulsification probability of the current mixed phase. The prediction probability is compared with the threshold to provide an early warning. The adjusted mixed phase is allowed to stand and separate into phases, and then back-extracted to obtain vanadium electrolyte.

2. The method for threshold control of vanadium electrolyte preparation based on emulsification probability prediction according to claim 1, characterized in that, The transient physical aging influencing factor obtained by mapping torque deviation level and ultrasonic offset level includes: First, the torque deviation level... and ultrasonic offset level Square each element separately and add them together; then multiply the sum of squares by a scaling factor obtained by fitting a historical dataset using the least squares method. Round the product to the nearest integer; finally, compare the rounded value with 10 and take the smaller one as the transient physical aging influence factor. Among them, torque deviation level and ultrasonic offset level The values ​​of all values ​​are integers from 0 to 4, representing the transient physical aging influencing factor. Integers from 0 to 10.

3. The method for threshold control in vanadium electrolyte preparation based on emulsification probability prediction according to claim 1, characterized in that, The time-series baseline drift index is obtained by summing the squares of the transient physical aging influencing factors from multiple consecutive batches using a time-sliding window weighted method, and then taking the square root. This includes: Recently The transient physical aging impact factors of each batch are stored in a time sliding window. The transient physical aging impact factors of each batch are squared and multiplied by the exponential decay weight. The sum of these factors is accumulated to obtain a weighted sum of squares, and then the square root is taken to obtain the time series baseline drift index.

4. The method for threshold control of vanadium electrolyte preparation based on emulsification probability prediction according to claim 1, characterized in that, The method for obtaining the intermediate compensation demand factor includes: calculating the ratio of the current cumulative operating time of the extractant to the upper limit of its rated life to obtain the extractant fatigue ratio. The fatigue ratio of the agitator is obtained by calculating the ratio of its cumulative running time to its design life limit. Through weighted sum The final fatigue ratio was obtained. ,in The ratio of the time-series baseline drift index to the final fatigue degree is used to determine the relationship between the time-series baseline drift index and the final fatigue degree. Multiply to obtain the intermediate compensation demand factor .

5. The method for threshold control in vanadium electrolyte preparation based on emulsification probability prediction according to claim 1, characterized in that, The process of converting intermediate compensation demand factors into compensation intensity coefficients using saturation mapping rules includes: If the intermediate compensation demand factor ,but ;like ,but ;like ,but ;like ,but ;like ,but ; This is the compensation strength coefficient.

6. The method for threshold control in vanadium electrolyte preparation based on emulsification probability prediction according to claim 1, characterized in that, The calculation of the threshold compensation multiplier based on the compensation intensity coefficient includes: ； In the formula, For threshold compensation multipliers; The maximum allowable threshold upward floating ratio, with a value range of [value range missing]. ; This is the compensation strength coefficient.

7. The method for threshold control of vanadium electrolyte preparation based on emulsification probability prediction according to claim 1, characterized in that, The random forest model includes: using stirring speed, phase volume ratio, system temperature, pH value, and the transient physical aging influencing factor as input feature vectors; during model inference, the decision threshold of decision tree nodes belonging to mechanical and fluid dynamic features in the random forest model is dynamically reconstructed using the threshold compensation multiplier: if the feature of the current decision tree node belongs to the mechanical and fluid dynamic feature list, the static partition threshold of the node is multiplied by the threshold compensation multiplier to obtain the dynamic threshold, and the dynamic threshold is used for branch judgment; if it belongs to the thermodynamic and chemical feature list, the static threshold remains unchanged; the mechanical and fluid dynamic feature list includes stirring speed, phase volume ratio, and transient physical aging influencing factor, and the thermodynamic and chemical feature list includes system temperature and pH value.

8. The method for threshold control in vanadium electrolyte preparation based on emulsification probability prediction according to claim 1, characterized in that, The random forest model is trained in the following way: Historical production data is used, with the emulsification state indicated by manual observation or by the organic phase content in the aqueous phase exceeding 5% after phase separation.

9. The method for threshold control in vanadium electrolyte preparation based on emulsification probability prediction according to claim 1, characterized in that, The method of issuing an early warning based on the comparison between the predicted emulsification probability and the threshold includes: issuing a red alarm when the emulsification probability is greater than 75%, and executing at least one of the following control commands: reducing the stirring speed by 5% each time for no more than three times, adjusting the volume ratio of the organic phase to the aqueous phase from 1:1 to 0.8:1, or stopping stirring and allowing it to stand for 60 minutes to break the emulsion.

10. A threshold control system for vanadium electrolyte preparation based on emulsification probability prediction, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the threshold control method for vanadium electrolyte preparation based on emulsification probability prediction according to any one of claims 1-9.

Images